Chemistry of Electronic Materials

1. Preface to the Chemistry of Electronic Materials
2. Background to Electronic Materials
1. Introduction to Semiconductors
2. Doped Semiconductors
3. Diffusion
4. Crystal Structure
5. Structures of Element and Compound Semiconductors
3. Device Fundamentals
1. Introduction to Bipolar Transistors
2. Basic MOS Structure
3. Introduction to the MOS Transistor and MOSFETs
4. Light Emitting Diode
5. Polymer Light Emitting Diodes
6. Laser
7. Solar Cells
4. Bulk Materials
1. Properties of Gallium Arsenide
2. Synthesis and Purification of Bulk Semiconductors
3. Growth of Gallium Arsenide Crystals
4. Ceramic Processing of Alumina
5. Piezoelectric Materials Synthesis
5. Wafer Formation and Processing
1. Formation of Silicon and Gallium Arsenide Wafers
. Doping
. Applications for Silica Thin Films
. Oxidation of Silicon
. Photolithography
. Composition and Photochemical Mechanisms of
Photoresists
8. Integrated Circuit Well and Gate Creation

NOUR WN

6. Thin Film Growth
1. Molecular Beam Epitaxy
2. Atomic Layer Deposition
3. Chemical Vapor Deposition
4. Liguid Phase Deposition
7. Chemical Vapor Deposition
1. Selecting a Molecular Precursor for Chemical Vapor
Deposition
2. Determination of Sublimation Enthalpy and Vapor
Pressure for Inorganic and Metal-Organic Compounds by
Thermogravimetric Analysis
3. 13-15 (III-V) Semiconductor Chemical Vapor Deposition
1. Phosphine and Arsine
2. Mechanism of the Metal Organic Chemical Vapor
Deposition of Gallium Arsenide
4. Oxide Chemical Vapor Deposition
1. Chemical Vapor Deposition of Silica Thin Films
2. Chemical Vapor Deposition of Alumina
5. Nitride Chemical Vapor Deposition
1. Introduction to Nitride Chemical Vapor Deposition
2. Chemical Vapor Deposition of Silicon Nitride and
Oxynitride
3. Chemical Vapor Deposition of Aluminum Nitride
6. Metal Organic Chemical Vapor Deposition of Calcium
Fluoride

8. Materials Characterization
1. Rutherford Backscattering of Thin Films
to the Study of Crystal Surface Processes .
3. Atomic Force Microscopy

9. Nanotechnology
1. Introduction to Nanoparticle Synthesis
2. Semiconductor Nanomaterials
1. Synthesis of Semiconductor Nanoparticles
2. Optical Properties of Group 12-16 (II-VI)
Semiconductor Nanoparticles
3. Characterization of Group 12-16 (II-VI)
Semiconductor Nanoparticles by UV-visible
Spectroscopy
Semiconductor Nanoparticles by Fluorescence
Spectroscopy
3. Carbon Nanomaterials
4. Graphene
5. Rolling Molecules on Surfaces Under STM Imaging
10. Economic and Environmental Issues
1. The Environmental Impact of the Manufacturing of
Seminconductors

Preface to the Chemistry of Electronic Materials

The intention of this text is not to provide a comprehensive reference to all
aspects of semiconductor device fabrication or a review of research results
that, irrespective of their promise, have not been adopted into mainstream
production. Instead it is aimed to provide a useful reference for those
interested in the chemical aspects of the electronics industry.

Given the nature of Connexions, this course is fluid in structure and
content. In addition, it contains modules by other authors where
appropriate. The content will be updated and expanded with time. If any
authors have suitable content, please contact me and I will be glad to assist
in transforming the content to a suitable module structure.

Andrew R. Barron

Rice University, Houston, TX 77005. E-mail: arb@rice.edu

Introduction to Semiconductors
Introduction to semiconductors, mainly looking at the behavior of electrons
in a solid from a quantum mechanical point of view.

Note:This module is adapted from the Connexions module entitled
Introduction to Semiconductors by Bill Wilson.

If we only had to worry about simple conductors, life would not be very
complicated, but on the other hand we wouldn't be able to make computers,
CD players, cell phones, i-Pods and a lot of other things which we have
found to be useful. We will now move on, and talk about another class of
conductors called semiconductors.

In order to understand semiconductors and in fact to get a more accurate
picture of how metals, or normal conductors actually work, we really have
to resort to quantum mechanics. Electrons in a solid are very tiny objects,
and it turns out that when things get small enough, they no longer exactly
following the classical "Newtonian" laws of physics that we are all familiar
with from everyday experience. It is not the purpose of this course to teach
quantum mechanics, so what we are going to do instead is describe the
results which come from looking at the behavior of electrons in a solid from
a quantum mechanical point of view.

Solids (at least the ones we will be talking about, and especially
semiconductors) are crystalline materials, which means that they have their
atoms arranged in a ordered fashion. We can take silicon (the most
important semiconductor) as an example. Silicon is a group 14(IV) element,
which means it has four electrons in its outer or valence shell. Silicon
crystallizes in a structure called the diamond crystal lattice, shown in [link].
Each silicon atom has four covalent bonds, arranged in a tetrahedral
formation about the atom center.

Crystal structure of
silicon.

In two dimensions, we can schematically represent a piece of single-crystal
silicon as shown in [link]. Each silicon atom shares its four valence
electrons with valence electrons from four nearest neighbors, filling the
shell to 8 electrons, and forming a stable, periodic structure. Once the atoms
have been arranged like this, the outer valence electrons are no longer
strongly bound to the host atom. The outer shells of all of the atoms blend
together and form what is called a band. The electrons are now free to move
about within this band, and this can lead to electrical conductivity as we
discussed earlier.

<4 —— st ———

—Si = —si—Ssi —* =—

—3 dd

A 2-D representation of a
silicon crystal.

This is not the complete story however, for it turns out that due to quantum
mechanical effects, there is not just one band which holds electrons, but
several of them. What will follow is a very qualitative picture of how the
electrons are distributed when they are in a periodic solid, and there are
necessarily some details which we will be forced to gloss over. On the other
hand this will give you a pretty good picture of what is going on, and may
enable you to have some understanding of how a semiconductor really
works. Electrons are not only distributed throughout the solid crystal
spatially, but they also have a distribution in energy as well. The potential
energy function within the solid is periodic in nature. This potential
function comes from the positively charged atomic nuclei which are
arranged in the crystal in a regular array. A detailed analysis of how
electron wave functions, the mathematical abstraction which one must use
to describe how small quantum mechanical objects behave when they are in
a periodic potential, gives rise to an energy distribution somewhat like that
shown in [link].

vi,
TF aoe
eee ais
Bo
ey eee ee

Band Gap

Schematic of the first two
bands in a periodic solid
showing energy levels
and bands.

Firstly, unlike the case for free electrons, in a periodic solid, electrons are
not free to take on any energy value they wish. They are forced into specific
energy levels called allowed states, which are represented by the cups in
[link]. The allowed states are not distributed uniformly in energy either.
They are grouped into specific configurations called energy bands. There
are no allowed levels at zero energy and for some distance above that.
Moving up from zero energy, we then encounter the first energy band. At
the bottom of the band there are very few allowed states, but as we move up
in energy, the number of allowed states first increases, and then falls off
again. We then come to a region with no allowed states, called an energy
band gap. Above the band gap, another band of allowed states exists. This
goes on and on, with any given material having many such bands and band
gaps. This situation is shown schematically in [link], where the small cups
represent allowed energy levels, and the vertical axis represents electron
energy.

It turns out that each band has exactly 2N allowed states in it, where N is the
total number of atoms in the particular crystal sample we are talking about.
(Since there are 10 cups in each band in the figure, it must represent a
crystal with just 5 atoms in it. Not a very big crystal at all!) Into these bands
we must now distribute all of the valence electrons associated with the
atoms, with the restriction that we can only put one electron into each
allowed state. This is the result of something called the Pauli exclusion
principle. Since in the case of silicon there are 4 valence electrons per atom,
we would just fill up the first two bands, and the next would be empty. If
we make the logical assumption that the electrons will fill in the levels with
the lowest energy first, and only go into higher lying levels if the ones
below are already filled. This situation is shown in [link], in which we have
represented electrons as small black balls with a "-" sign on them. Indeed,
the first two bands are completely full, and the next is empty. What will
happen if we apply an electric field to the sample of silicon? Remember the
diagram we have at hand right now is an energy based one, we are showing
how the electrons are distributed in energy, not how they are arranged
spatially. On this diagram we can not show how they will move about, but
only how they will change their energy as a result of the applied field. The

electric field will exert a force on the electrons and attempt to accelerate
them. If the electrons are accelerated, then they must increase their kinetic
energy. Unfortunately, there are no empty allowed states in either of the
filled bands. An electron would have to jump all the way up into the next
(empty) band in order to take on more energy. In silicon, the gap between
the top of the highest most occupied band and the lowest unoccupied band
is 1.1 eV. (One eV is the potential energy gained by an electron moving
across an electrical potential of one volt.) The mean free path or distance
over which an electron would normally move before it suffers a collision is
only a few hundred angstroms (ca. 300 x 10°8 cm) and so you would need a
very large electric field (several hundred thousand V/cm) in order for the
electron to pick up enough energy to "jump the gap". This makes it appear
that silicon would be a very bad conductor of electricity, and in fact, very
pure silicon is very poor electrical conductor.

VAVAVAY!
WAU
vi

bands
full

and the
next

empty.

A metal is an element with an odd number of valence electrons so that a
metal ends up with an upper band which is just half full of electrons. This is
illustrated in [link]. Here we see that one band is full, and the next is just
half full. This would be the situation for the Group 13(III) element
aluminum for instance. If we apply an electric field to these carriers, those
near the top of the distribution can indeed move into higher energy levels
by acquiring some kinetic energy of motion, and easily move from one
place to the next. In reality, the whole situation is a bit more complex than
we have shown here, but this is not too far from how it actually works.

vi
whe
VATAVAY!
wavs
ws

Electron
distributio

nfora
metal or
good
conductor.

So, back to our silicon sample. If there are no places for electrons to "move'
into, then how does silicon work as a "Semiconductor"? Well, in the first
place, it turns out that not all of the electrons are in the bottom two bands.
In silicon, unlike say quartz or diamond, the band gap between the top-most
full band, the next empty one is not so large. As we mentioned above it is
only about 1.1 eV. So long as the silicon is not at absolute zero temperature,
some electrons near the top of the full band can acquire enough thermal
energy that they can "hop" the gap, and end up in the upper band, called the
conduction band. This situation is shown in [link].

UU?

Thermal
excitation
of
electrons
across the
band gap.

In silicon at room temperature, roughly 10!” electrons per cubic centimeter
are thermally excited across the band-gap at any one time. It should be
noted that the excitation process is a continuous one. Electrons are being
excited across the band, but then they fall back down into empty spots in
the lower band. On average however, the 10!° in each cm? of silicon is what
you will find at any given instant. Now 10 billion electrons per cubic
centimeter seems like a lot of electrons, but lets do a simple calculation.
The mobility of electrons in silicon is about 1000 cm?/V.s. Remember,
mobility times electric field yields the average velocity of the carriers.
Electric field has units of V/cm, so with these units we get velocity in cm/s
as we should. The charge on an electron is 1.6 x 10°! coulombs. Thus from
[link]:

Equation:

o = nqu
= 10!° (1.6 x 10°!) 1000
= 1.6x 10° mhos/em

If we have a sample of silicon 1 cm long by (1 mm x 1mm) square, it would
have a resistance, [link], which does not make it much of a "conductor". In
fact, if this were all there was to the silicon story, we could pack up and
move on, because at any reasonable temperature, silicon would conduct
electricity very poorly.

Equation:

R =L/oA
1/(1.6 x 10°°)(0.1)?
1.6 x 10° MQ

Doped Semiconductors
From the silicon's crystal structure to discuss how to make doped
semiconductors and the mechanics.

Note:This module is adapted from the Connexions module entitled Doped
Semiconductors by Bill Wilson.

To see how we can make silicon a useful electronic material, we will have
to go back to its crystal structure ([link]). Suppose somehow we could
substitute a few atoms of phosphorus for some of the silicon atoms.

L | | -
— Si —Ssi—=si=si =SsSi7
I IE tl I
eS |S ee nd | oe — e
IY WT
—_ Si — P=—Sj ——Si ——Si —

A two dimensional
representation of a
silicon crystal lattice
"doped" with
phosphorus.

If you sneak a look at the periodic table, you will see that phosphorus is a
group V element, as compared with silicon which is a group 14(IV)
element. What this means is the phosphorus atom has five outer or valence
electrons, instead of the four which silicon has. In a lattice composed
mainly of silicon, the extra electron associated with the phosphorus atom
has no "mating" electron with which it can complete a shell, and so is left
loosely dangling to the phosphorus atom, with relatively low binding
energy. In fact, with the addition of just a little thermal energy (from the

natural or latent heat of the crystal lattice) this electron can break free and
be left to wander around the silicon atom freely. In our "energy band"
picture, we have something like what we see in [link]. The phosphorus
atoms are represented by the added cups with P's on them. They are new
allowed energy levels which are formed within the "band gap" near the
bottom of the first empty band. They are located close enough to the empty
(or "conduction") band, so that the electrons which they contain are easily
excited up into the conduction band. There, they are free to move about and
contribute to the electrical conductivity of the sample. Note also, however,
that since the electron has left the vicinity of the phosphorus atom, there is
now one more proton than there are electrons at the atom, and hence it has a
net positive charge of 1q. We have represented this by putting a little "+"
sign in each P-cup. Note that this positive charge is fixed at the site of the
phosphorous atom called a donor since it "donates" an electron up into the
conduction band, and is not free to move about in the crystal.

y y \ Conduction
(ee

# Y ¥ ai Band Gap

Ue i lvage
wAwi Band

Silicon doped with
phosphorus.

How many phosphorus atoms do we need to significantly change the
resistance of our silicon? Suppose we wanted our 1 mm by 1 mm square
sample to have a resistance of one ohm as opposed to more than 60 MQ.

Turning the resistance equation around we get, [link]. And hence, if we
continue to assume an electron mobility of 1000 cm?/volt.sec, [link].
Equation:

o = L/RA

= 1Q/1 x (0.1)

= 100 mho/cm
Equation:
n = O/qu

100/(1.6 x 10°!°)1000
= 6.25 x 10!7 cm?

Now adding more than 6 x 10!” phosphorus atoms per cubic centimeter
might seem like a lot of phosphorus, until you realize that there are almost
10*4 silicon atoms in a cubic centimeter and hence only one in every 1.6
million silicon atoms has to be changed into a phosphorus one to reduce the
resistance of the sample from several 10s of MQ down to only one Q. This
is the real power of semiconductors. You can make dramatic changes in
their electrical properties by the addition of only minute amounts of
impurities. This process is called doping the semiconductor. It is also one of
the great challenges of the semiconductor manufacturing industry, for it is
necessary to maintain fantastic levels of control of the impurities in the
material in order to predict and control their electrical properties.

Again, if this were the end of the story, we still would not have any
calculators, cell phones, or stereos, or at least they would be very big and
cumbersome and unreliable, because they would have to work using
vacuum tubes. We now have to focus on the few "empty" spots in the lower,
almost full band (called the valence band.) We will take another view of
this band, from a somewhat different perspective. I must confess at this
point that what I am giving you is even further from the way things really
work, then the "cups at different energies" picture we have been using so
far. The problem is, that in order to do things right, we have to get involved
in momentum phase-space, a lot more quantum mechanics, and generally a
bunch of math and concepts we don't really need in order to have some idea

of how semiconductor devices work. What follow below is really intended
as a motivation, so that you will have some feeling that what we state as
results, is actually reasonable.

Consider [link]. Here we show all of the electrons in the valence, or almost
full band, and for simplicity show one missing electron. Let's apply an
electric field, as shown by the arrow in the figure. The field will try to move
the (negatively charged) electrons to the left, but since the band is almost
completely full, the only one that can move is the one right next to the
ony spot, or hole as it is called.

Sn

Band full of electrons,
with one missing.

One thing you may be worrying about is what happens to the electrons at
the ends of the sample. This is one of the places where we are getting a
somewhat distorted view of things, because we should really be looking in
momentum, or wave-vector space rather than "real" space. In that picture,
they magically drop off one side and "reappear" on the other. This doesn't
happen in real space of course, so there is no easy way we can deal with it.

A short time after we apply the electric field we have the situation shown in
[link], and a little while after that we have [link]. We can interpret this
motion in two ways. One is that we have a net flow of negative charge to
the left, or if we consider the effect of the aggregate of all the electrons in
the band we could picture what is going on as a single positive charge,
moving to the right. This is shown in [link]. Note that in either view we
have the same net effect in the way the total net charge is transported

through the sample. In the mostly negative charge picture, we have a net
flow of negative charge to the left. In the single positive charge picture, we
have a net flow of positive charge to the right. Both give the same sign for
the current!

E

Motion of the
"missing" electron
with an electric field.

Further motion of the
"missing electron"
spot.

Motion of a "hole"
due to an applied
electric field.

Thus, it turns out, we can consider the consequences of the empty spaces
moving through the co-ordinated motion of electrons in an almost full band
as being the motion of positive charges, moving wherever these empty
spaces happen to be. We call these charge carriers "holes" and they too can
add to the total conduction of electricity in a semiconductor. Using p to
represent the density (in cm” of spaces in the valence band and py, and pip, to
represent the mobility of electrons and holes respectively (they are usually
not the same) we can modify to give the conductivity 0, when both
electrons’ holes are present, [link].

Equation:

Oo = nqu, + Pq,

How can we get a sample of semiconductor with a lot of holes in it? What
if, instead of phosphorus, we dope our silicon sample with a group III
element, say boron? This is shown in [link]. Now we have some missing
orbitals, or places where electrons could go if they were around. This
modifies our energy picture as follows in [link]. Now we see a set of new
levels introduced by the boron atoms. They are located within the band gap,
just a little way above the top of the almost full, or valence band. Electrons
in the valence band can be thermally excited up into these new allowed
levels, creating empty states, or holes, in the valence band. The excited

electrons are stuck at the boron atom sites called acceptors, since they
"accept" an electron from the valence band, and hence act as fixed negative
charges, localized there. A semiconductor which is doped predominantly
with acceptors is called p-type, and most of the electrical conduction takes
place through the motion of holes. A semiconductor which is doped with
donors is called n-type, and conduction takes place mainly through the
motion of electrons.

a ee ee ey
— sol ol —— ool
ott tl oH |
<=B-— sisi B= Si-—
(Ey | | |
—=Si= BB Si =si si.

A two dimensional
representation of a
silicon crystal lattice
doped with boron.

UB
oF ee!

@ e up Band Gap
a Valence
Band

P-type silicon, due to
boron acceptors.

In n-type material, we can assume that all of the phosphorous atoms, or
donors, are fully ionized when they are present in the silicon structure.
Since the number of donors is usually much greater than the native, or
intrinsic electron concentration, (* 10'° cm”), if Ny is the density of donors
in the material, then n, the electron concentration, ~ Ng. If an electron
deficient material such as boron is present, then the material is called p-type
silicon, and the hole concentration is just * N, the concentration of
acceptors, since these atoms "accept" electrons from the valence band.

If both donors and acceptors are in the material, then which ever one has the
higher concentration wins out. This is called compensation. If there are
more donors than acceptors then the material is n-type and n * Nj, - Ng. If
there are more acceptors than donors then the material is p-type and p * N,
- Nq. It should be noted that in most compensated material, one type of
impurity usually has a much greater (several order of magnitude)
concentration than the other, and so the subtraction process described above
usually does not change things very much, e.g., 10/8 - 1016 = 1018.

One other fact which you might find useful is that, again, because of
quantum mechanics, it turns out that the product of the electron and hole
concentration in a material must remain a constant. In silicon at room
temperature:

Equation:

Thus, if we have an n-type sample of silicon doped with 10!” donors per
cubic centimeter, then n, the electron concentration is just p , the hole
concentration, is 102°/10!” = 10° cm’. The carriers which dominate a
material are called majority carriers, which would be the electrons in the
above example. The other carriers are called minority carriers (the holes in
the example) and while 10° might not seem like much compared to 10!” the
presence of minority carriers is still quite important and can not be ignored.
Note that if the material is undoped, then it must be that n = p and n = p=
10"

The picture of "cups" of different allowed energy levels is useful for
gaining a pictorial understanding of what is going on in a semiconductor,
but becomes somewhat awkward when you want to start looking at devices
which are made up of both n and p type silicon. Thus, we will introduce one
more way of describing what is going on in our material. The picture shown
in [link] is called a band diagram. A band diagram is just a representation
of the energy as a function of position with a semiconductor device. In a
band diagram, positive energy for electrons is upward, while for holes,
positive energy is downwards. That is, if an electron moves upward, its
potential energy increases just as a with a normal mass in a gravitational
field. Also, just as a mass will "fall down" if given a chance, an electron
will move down a slope shown in a band diagram. On the other hand, holes
gain energy by moving downward and so they have a tendancy to "float"
upward if given the chance - much like a bubble in a liquid. The line
labeled E,, in [link] shows the edge of the conduction band, or the bottom of
the lowest unoccupied allowed band, while E,, is the top edge of the
valence, or highest occupied band. The band gap, E, for the material is
obviously E, - Ey. The dotted line labeled E; is called the Fermi level and it
tells us something about the chemical equilibrium energy of the material,
and also something about the type and number of carriers in the material.
More on this later. Note that there is no zero energy level on a diagram such
as this. We often use either the Fermi level or one or other of the band edges
as a reference level on lieu of knowing exactly where "zero energy" is

located.
Energy (eV)
Ec
———— Ej
Ey
Position

An electron band-
diagram for a
semiconductor.

The distance (in energy) between the Fermi level and either E,, and E,, gives
us information concerning the density of electrons and holes in that region
of the semiconductor material. The details, once again, will have to be
begged off on grounds of mathematical complexity. It turns out that you can
Say:

Equation:

Equation:

kT

p=Ne

Both N, and N,, are constants that depend on the material you are talking
about, but are typically on the order of 10!9 cm’. The expression in the
denominator of the exponential is just Boltzman's constant (8.63 x 10°
eV/K), k, times the temperature T of the material (in absolute temperature
or Kelvin). At room temperature kT = '/49 of an electron volt. Look
carefully at the numerators in the exponential. Note first that there is a
minus sign in front, which means the bigger the number in the exponent, the
fewer carriers we have. Thus, the top expression says that if we have n-type
material, then E must not be too far away from the conduction band, while
if we have p-type material, then the Fermi level,E, must be down close to
the valence band. The closer EF gets to E, the more electrons we have. The
closer Ey gets to Ey, the more holes we have. [link] therefore must be for a
sample of n-type material. Note also that if we know how heavily a sample
is doped (i.e., we know what Nj is) and from the fact that n * Ng we can use
[link] to find out how far away the Fermi level is from the conduction band,
[link].

Equation:

N.

To help further in our ability to picture what is going on, we will often add
to this band diagram, some small signed circles to indicate the presence of
mobile electrons and holes in the material. Note that the electrons are
spread out in energy. From our "cups" picture we know they like to stay in
the lower energy states if possible, but some will be distributed into the
higher levels as well. What is distorted here is the scale. The band-gap for
silicon is 1.1 eV, while the actual spread of the electrons would probably
only be a few tenths of an eV, not nearly as much as is shown in [link]. Lets
look at a sample of p-type material, just for comparison. Note that for holes,
increasing energy goes down not up, so their distribution is inverted from
that of the electrons. You can kind of think of holes as bubbles in a glass of
soda or beer, they want to float to the top if they can. Note also for both n
and p-type material there are also a few "minority" carriers, or carriers of
the opposite type, which arise from thermal generation across the band-gap.

© ©
CROMOMC
CIOMONOKCRONS

© ©

Band diagram for an

n-type
semiconductor.

Diffusion
The module discusses the process of electrons moving across a p-n or n-p
junction known as diffusion.

Note:This module is adapted from the Connexions module entitled
Diffusion by Bill Wilson.

Introduction

Let us turn our attention to what happens to the electrons and holes once
they have been injected across a forward-biased junction. We will
concentrate just on the electrons which are injected into the p-side of the
junction, but keep in mind that similar things are also happening to the
holes which enter the n-side.

When electrons are injected across a junction, they move away from the
junction region by a diffusion process, while at the same time, some of
them are disappearing because they are minority carriers (electrons in
basically p-type material) and so there are lots of holes around for them to
recombine with. This is all shown schematically in [link].

Injection

Processes involved in
electron transport
across a p-n junction.

Diffusion process quantified

It is actually fairly easy to quantify this, and come up with an expression for
the electron distribution within the p-region. First we have to look a little bit
at the diffusion process however. Imagine that we have a series of bins,
each with a different number of electrons in them. In a given time, we could
imagine that all of the electrons would flow out of their bins into the
neighboring ones. Since there is no reason to expect the electrons to favor
one side over the other, we will assume that exactly half leave by each side.
This is all shown in [link]. We will keep things simple and only look at
three bins. Imagine there are 4, 6, and 8 electrons respectively in each of the
bins. After the required "emptying time," we will have a net flux of exactly
one electron across each boundary as shown.

A schematic
representation of a
diffusion problem.

Now let's raise the number of electrons to 8, 12 and 16 respectively ({link]).
We find that the net flux across each boundary is now 2 electrons per
emptying time, rather than one. Note that the gradient (slope) of the
concentration in the boxes has also doubled from one per box to two per
box. This leads us to a rather obvious statement that the flux of carriers is
proportional to the gradient of their density. This is stated formally in what

is known as Fick's First Law of Diffusion, [link]. Where D, is simply a
proportionality constant called the diffusion coefficient. Since we are
talking about the motion of electrons, this diffusion flux must give rise to a
current density J.,.... Since an electron has a charge —q associated with it,
[link].

Equation:
d
een a oe ee
d x
Equation:
dn
ease = We au

A schematic
representation of a
diffusion from bins.

Now we have to invoke something called the continuity equation. Imagine
we have a volume (V) which is filled with some charge (Q). It is fairly
obvious that if we add up all of the current density which is flowing out of
the volume that it must be equal to the time rate of decrease of the charge
within that volume. This ideas is expressed in the formula below which uses
a closed-surface integral, along with the all the other integrals to follow:
Equation:

We can write @ as, [link], where we are doing a volume integral of the
charge density (p ) over the volume (V). Now we can use Gauss' theorem
which says we can replace a surface integral of a quantity with a volume
integral of its divergence, [link].

Equation:

Q= $v) av

V

fras= | aiv(aav

S

Equation:

So, combining [link], [link] and [link], we have, [link].
Equation:
dp
div (J) dV =— | —dV
/ iv (J) / dt

Finally, we let the volume V shrink down to a point, which means the
quantities inside the integral must be equal, and we have the differential
form of the continuity equation (in one dimension), [Link].

Equation:

div (J) os

What about the electrons?

Now let's go back to the electrons in the diode. The electrons which have
been injected across the junction are called excess minority carriers,
because they are electrons in a p-region (hence minority) but their
concentration is greater than what they would be if they were in a sample of
p-type material at equilibrium. We will designate them as n', and since they
could change with both time and position we shall write them as n'(x,¢).
Now there are two ways in which n'(x,t) can change with time. One would
be if we were to stop injecting electrons in from the n-side of the junction.
A reasonable way to account for the decay which would occur if we were
not supplying electrons would be to write:

Equation:

Where 7, called the minority carrier recombination lifetime. It is pretty easy
to show that if we start out with an excess minority carrier concentration no'
at t = 0, then n'(x,t) will go as, [link]. But, the electron concentration can
also change because of electrons flowing into or out of the region x. The

p(x,t)
q

electron concentration n'(x,t) is just . Thus, due to electron flow we

have, [Link].

Equation:
n'(x,t) = n'jem
Equation:
n'(z,t) = a sole.t)
= a div (J(z,t))

But, we can get an expression for J(, t) from [link]. Reducing the
divergence in [link] to one dimension (we just have a oh) we finally end up

with, [Link].
Equation:

d? n'(z, t)
d x?

Combining [link] and [link] (electrons will, after all, suffer from both
recombination and diffusion) and we end up with:
Equation:

d? n'(z, t) n' (a, t)
/ ? ’
—n (x,t) = De = 5 = :

This is a somewhat specialized form of an equation called the ambipolar
diffusion equation. It seems kind of complicated but we can get some nice
results from it if we make some simply boundary condition assumptions.

Using the ambipolar diffusion equation

For anything we will be interested in, we will only look at steady state
solutions. This means that the time derivative on the LHS of [link] is zero,
and so letting n’(x,t) become simply n‘(a) since we no longer have any
time variation to worry about, we have:

Equation:

d? 1
Day

n'(x) =0

Picking the not unreasonable boundary conditions that n’(0) = no (the
concentration of excess electrons just at the start of the diffusion region)
and n'(a) —> 0. as 2 — oo (the excess carriers go to zero when we get far
from the junction) then:

Equation:

The expression in the radical ./ D.7T, is called the electron diffusion length,
L,, and gives us some idea as to how far away from the junction the excess
electrons will exist before they have more or less all recombined. This will
be important for us when we move on to bipolar transistors. A typical value
for the diffusion coefficient for electrons in silicon would be D, = 25
cm?/sec and the minority carrier lifetime is usually around a microsecond.
As shown in [link] this is not very far at all.

Equation:

Le JV Det
— 4/25 x 10-6

— 5x10°cm

Crystal Structure

Introduction

In any sort of discussion of crystalline materials, it is useful to begin with a
discussion of crystallography: the study of the formation, structure, and
properties of crystals. A crystal structure is defined as the particular
repeating arrangement of atoms (molecules or ions) throughout a crystal.
Structure refers to the internal arrangement of particles and not the external
appearance of the crystal. However, these are not entirely independent since
the external appearance of a crystal is often related to the internal
arrangement. For example, crystals of cubic rock salt (NaCl) are physically
cubic in appearance. Only a few of the possible crystal structures are of
concern with respect to simple inorganic salts and these will be discussed in
detail, however, it is important to understand the nomenclature of
crystallography.

Crystallography

Bravais lattice

The Bravais lattice is the basic building block from which all crystals can
be constructed. The concept originated as a topological problem of finding
the number of different ways to arrange points in space where each point
would have an identical “atmosphere”. That is each point would be
surrounded by an identical set of points as any other point, so that all points
would be indistinguishable from each other. Mathematician Auguste
Bravais discovered that there were 14 different collections of the groups of
points, which are known as Bravais lattices. These lattices fall into seven
different "crystal systems”, as differentiated by the relationship between the
angles between sides of the “unit cell” and the distance between points in
the unit cell. The unit cell is the smallest group of atoms, ions or molecules
that, when repeated at regular intervals in three dimensions, will produce
the lattice of a crystal system. The “lattice parameter” is the length between
two points on the comers of a unit cell. Each of the various lattice
parameters are designated by the letters a, b, and c. If two sides are equal,

such as in a tetragonal lattice, then the lengths of the two lattice parameters
are designated a and c, with b omitted. The angles are designated by the
Greek letters a, B, and y, such that an angle with a specific Greek letter is
not subtended by the axis with its Roman equivalent. For example, a is the
included angle between the b and c axis.

[link] shows the various crystal systems, while [link] shows the 14 Bravais
lattices. It is important to distinguish the characteristics of each of the
individual systems. An example of a material that takes on each of the
Bravais lattices is shown in [link].

System Axial lengths and angles Parcel
geometry

cubic a=b=c,a= B= y= 90°

tetragonal a=b#c,a=fB=~7y=90°

orthorhombic a#b#c,a= 68 = y= 90°

rhombohedral a=b=c,a=B=y77#90°

a=b4#c,a=B=90°, y=

h ]
exagona 120°
monoclinic ae ARE
90°
triclinic a#%~b#c,azpFy

Geometrical characteristics of the seven crystal systems.

simple cubic body-centered face-centered
cubic cubic

=
C

le
|
-»
i
simple body-centered
tetragonal tetragonal

simple body-centered
orthorhombic orthorhombic

base-centered face-centered
orthorhombic orthorhombic

rhombohedral hexagonal
simple base-centered triclinic
monoclinic monoclinic

Bravais lattices.

Crystal system Example

triclinic K S208

monoclinic As,S4, KNO>
rhombohedral Hg, Sb
hexagonal Zn, Co, NiAs
orthorhombic Ga, Fe3C
tetragonal In, TiO
cubic Au, Si, NaCl

Examples of elements and compounds that adopt each of the crystal
systems.

The cubic lattice is the most symmetrical of the systems. All the angles are
equal to 90°, and all the sides are of the same length (a = b = c). Only the
length of one of the sides (a) is required to describe this system completely.
In addition to simple cubic, the cubic lattice also includes body-centered
cubic and face-centered cubic ([{link]). Body-centered cubic results from the
presence of an atom (or ion) in the center of a cube, in addition to the atoms
(ions) positioned at the vertices of the cube. In a similar manner, a face-
centered cubic requires, in addition to the atoms (ions) positioned at the
vertices of the cube, the presence of atoms (ions) in the center of each of the
cubes face.

The tetragonal lattice has all of its angles equal to 90°, and has two out of
the three sides of equal length (a = b). The system also includes body-
centered tetragonal ([link]).

In an orthorhombic lattice all of the angles are equal to 90°, while all of its
sides are of unequal length. The system needs only to be described by three
lattice parameters. This system also includes body-centered orthorhombic,
base-centered orthorhombic, and face-centered orthorhombic ([link]). A
base-centered lattice has, in addition to the atoms (ions) positioned at the

vertices of the orthorhombic lattice, atoms (ions) positioned on just two
opposing faces.

The rhombohedral lattice is also known as trigonal, and has no angles equal
to 90°, but all sides are of equal length (a = b = c), thus requiring only by
one lattice parameter, and all three angles are equal (a = B = 4).

A hexagonal crystal structure has two angles equal to 90°, with the other
angle ( y) equal to 120°. For this to happen, the two sides surrounding the
120° angle must be equal (a = b), while the third side (c) is at 90° to the
other sides and can be of any length.

The monoclinic lattice has no sides of equal length, but two of the angles
are equal to 90°, with the other angle (usually defined as B) being
something other than 90°. It is a tilted parallelogram prism with rectangular
bases. This system also includes base-centered monoclinic ([link]).

In the triclinic lattice none of the sides of the unit cell are equal, and none of
the angles within the unit cell are equal to 90°. The triclinic lattice is chosen
such that all the internal angles are either acute or obtuse. This crystal
system has the lowest symmetry and must be described by 3 lattice
parameters (a, b, and c) and the 3 angles (a, B, and ¥).

Atom positions, crystal directions and Miller indices

Atom positions and crystal axes

The structure of a crystal is defined with respect to a unit cell. As the entire
crystal consists of repeating unit cells, this definition is sufficient to
represent the entire crystal. Within the unit cell, the atomic arrangement is
expressed using coordinates. There are two systems of coordinates
commonly in use, which can cause some confusion. Both use a corner of
the unit cell as their origin. The first, less-commonly seen system is that of
Cartesian or orthogonal coordinates (X, Y, Z). These usually have the units
of Angstroms and relate to the distance in each direction between the origin

of the cell and the atom. These coordinates may be manipulated in the same
fashion are used with two- or three-dimensional graphs. It is very simple,
therefore, to calculate inter-atomic distances and angles given the Cartesian
coordinates of the atoms. Unfortunately, the repeating nature of a crystal
cannot be expressed easily using such coordinates. For example, consider a
cubic cell of dimension 3.52 A. Pretend that this cell contains an atom that
has the coordinates (1.5, 2.1, 2.4). That is, the atom is 1.5 A away from the
origin in the x direction (which coincides with the a cell axis), 2.1 A in the
y (which coincides with the b cell axis) and 2.4 A in the z (which coincides
with the c cell axis). There will be an equivalent atom in the next unit cell
along the x-direction, which will have the coordinates (1.5 + 3.52, 2.1, 2.4)
or (5.02, 2.1, 2.4). This was a rather simple calculation, as the cell has very
high symmetry and so the cell axes, a, b and c, coincide with the Cartesian
axes, X, Y and Z. However, consider lower symmetry cells such as triclinic
or monoclinic in which the cell axes are not mutually orthogonal. In such
cases, expressing the repeating nature of the crystal is much more difficult
to accomplish.

Accordingly, atomic coordinates are usually expressed in terms of fractional
coordinates, (x, y, z). This coordinate system is coincident with the cell axes
(a, b, c) and relates to the position of the atom in terms of the fraction along
each axis. Consider the atom in the cubic cell discussion above. The atom
was 1.5 A in the a direction away from the origin. As the a axis is 3.52 A
long, the atom is (17/352) or 0.43 of the axis away from the origin.
Similarly, it is (*/3.59) or 0.60 of the b axis and (74/3) or 0.68 of the c axis.
The fractional coordinates of this atom are, therefore, (0.43, 0.60, 0.68).
The coordinates of the equivalent atom in the next cell over in the a
direction, however, are easily calculated as this atom is simply 1 unit cell
away ina. Thus, all one has to do is add 1 to the x coordinate: (1.43, 0.60,
0.68). Such transformations can be performed regardless of the shape of the
unit cell. Fractional coordinates, therefore, are used to retain and manipulate
crystal information.

Crystal directions

The designation of the individual vectors within any given crystal lattice is
accomplished by the use of whole number multipliers of the lattice
parameter of the point at which the vector exits the unit cell. The vector is
indicated by the notation [hkl], where h, k, and ! are reciprocals of the point
at which the vector exits the unit cell. The origination of all vectors is
assumed defined as [000]. For example, the direction along the a-axis
according to this scheme would be [100] because this has a component only
in the a-direction and no component along either the b or c axial direction.
A vector diagonally along the face defined by the a and b axis would be
[110], while going from one corner of the unit cell to the opposite corner
would be in the [111] direction. [link] shows some examples of the various
directions in the unit cell. The crystal direction notation is made up of the
lowest combination of integers and represents unit distances rather than
actual distances. A [222] direction is identical to a [111], so [111] is used.
Fractions are not used. For example, a vector that intercepts the center of
the top face of the unit cell has the coordinates x = 1/2, y = 1/2,z = 1. All
have to be inversed to convert to the lowest combination of integers (whole
numbers); i.e., [221] in [link]. Finally, all parallel vectors have the same
crystal direction, e.g., the four vertical edges of the cell shown in [link] all
have the crystal direction [hk/] = [001].

Some common
directions in a
cubic unit cell.

Crystal directions may be grouped in families. To avoid confusion there
exists a convention in the choice of brackets surrounding the three numbers
to differentiate a crystal direction from a family of direction. For a
direction, square brackets [hkl] are used to indicate an individual direction.
Angle brackets <hkl> indicate a family of directions. A family of directions
includes any directions that are equivalent in length and types of atoms
encountered. For example, in a cubic lattice, the [100], [010], and [001]
directions all belong to the <100> family of planes because they are
equivalent. If the cubic lattice were rotated 90°, the a, b, and c directions
would remain indistinguishable, and there would be no way of telling on
which crystallographic positions the atoms are situated, so the family of
directions is the same. In a hexagonal crystal, however, this is not the case,
so the [100] and [010] would both be <100> directions, but the [001]
direction would be distinct. Finally, negative directions are identified with a
bar over the negative number instead of a minus sign.

Crystal planes

Planes in a crystal can be specified using a notation called Miller indices.
The Miller index is indicated by the notation [hkl] where h, k, and | are
reciprocals of the plane with the x, y, and z axes. To obtain the Miller
indices of a given plane requires the following steps:

The plane in question is placed on a unit cell.

Its intercepts with each of the crystal axes are then found.

The reciprocal of the intercepts are taken.

These are multiplied by a scalar to insure that is in the simple ratio of whole
numbers.

For example, the face of a lattice that does not intersect the y or z axis
would be (100), while a plane along the body diagonal would be the (111)
plane. An illustration of this along with the (111) and (110) planes is given
in [Link].

1 _
Tepe = M10)

Examples of Miller indices
notation for crystal planes.

As with crystal directions, Miller indices directions may be grouped in
families. Individual Miller indices are given in parentheses (hkl), while
braces {hkl} are placed around the indices of a family of planes. For
example, (001), (100), and (010) are all in the {100} family of planes, for a
cubic lattice.

Description of crystal structures

Crystal structures may be described in a number of ways. The most
common manner is to refer to the size and shape of the unit cell and the
positions of the atoms (or ions) within the cell. However, this information is
sometimes insufficient to allow for an understanding of the true structure in
three dimensions. Consideration of several unit cells, the arrangement of the

atoms with respect to each other, the number of other atoms they in contact
with, and the distances to neighboring atoms, often will provide a better
understanding. A number of methods are available to describe extended
solid-state structures. The most applicable with regard to elemental and
compound semiconductor, metals and the majority of insulators is the close
packing approach.

Close packed structures: hexagonal close packing and cubic close
packing

Many crystal structures can be described using the concept of close
packing. This concept requires that the atoms (ions) are arranged so as to
have the maximum density. In order to understand close packing in three
dimensions, the most efficient way for equal sized spheres to be packed in
two dimensions must be considered.

The most efficient way for equal sized spheres to be packed in two
dimensions is shown in [link], in which it can be seen that each sphere (the
dark gray shaded sphere) is surrounded by, and is in contact with, six other
spheres (the light gray spheres in [link]). It should be noted that contact
with six other spheres the maximum possible is the spheres are the same
size, although lower density packing is possible. Close packed layers are
formed by repetition to an infinite sheet. Within these close packed layers,
three close packed rows are present, shown by the dashed lines in [link].

Schematic representation of a
close packed layer of equal
sized spheres. The close packed
rows (directions) are shown by
the dashed lines.

The most efficient way for equal sized spheres to be packed in three
dimensions is to stack close packed layers on top of each other to give a
close packed structure. There are two simple ways in which this can be
done, resulting in either a hexagonal or cubic close packed structures.

Hexagonal close packed

If two close packed layers A and B are placed in contact with each other so

as to maximize the density, then the spheres of layer B will rest in the
hollow (vacancy) between three of the spheres in layer A. This is
demonstrated in [link]. Atoms in the second layer, B (shaded light gray),
may occupy one of two possible positions ([link]a or b) but not both
together or a mixture of each. If a third layer is placed on top of layer B
such that it exactly covers layer A, subsequent placement of layers will

result in the following sequence ...ABABAB.... This is known as hexagonal

close packing or hcp.

(a) (b)

Schematic representation of two close packed
layers arranged in A (dark grey) and B (light grey)
positions. The alternative stacking of the B layer is

shown in (a) and (b).

The hexagonal close packed cell is a derivative of the hexagonal Bravais
lattice system ({link]) with the addition of an atom inside the unit cell at the
coordinates (1/3,7/3,'/9). The basal plane of the unit cell coincides with the
close packed layers ({link]). In other words the close packed layer makes-up
the {001} family of crystal planes.

A schematic
projection of the
basal plane of the
hep unit cell on the
close packed
layers.

The “packing fraction” in a hexagonal close packed cell is 74.05%; that is
74.05% of the total volume is occupied. The packing fraction or density is
derived by assuming that each atom is a hard sphere in contact with its
nearest neighbors. Determination of the packing fraction is accomplished
by calculating the number of whole spheres per unit cell (2 in hcp), the
volume occupied by these spheres, and a comparison with the total volume
of a unit cell. The number gives an idea of how “open” or filled a structure
is. By comparison, the packing fraction for body-centered cubic ({link]) is
68% and for diamond cubic (an important semiconductor structure to be
described later) is it 34%.

Cubic close packed: face-centered cubic

In a similar manner to the generation of the hexagonal close packed
structure, two close packed layers are stacked ([link]) however, the third
layer (C) is placed such that it does not exactly cover layer A, while sitting
in a set of troughs in layer B ([link]), then upon repetition the packing
sequence will be .. ABCABCABC.... This is known as cubic close packing
or ccp.

Schematic representation of the
three close packed layers in a
cubic close packed
arrangement: A (dark grey), B

(medium grey), and C (light
grey).

The unit cell of cubic close packed structure is actually that of a face-
centered cubic (fcc) Bravais lattice. In the fcc lattice the close packed layers
constitute the {111} planes. As with the hcp lattice packing fraction in a
cubic close packed (fcc) cell is 74.05%. Since face centered cubic or fcc is
more commonly used in preference to cubic close packed (ccp) in
describing the structures, the former will be used throughout this text.

Coordination number

The coordination number of an atom or ion within an extended structure is
defined as the number of nearest neighbor atoms (ions of opposite charge)
that are in contact with it. A slightly different definition is often used for
atoms within individual molecules: the number of donor atoms associated
with the central atom or ion. However, this distinction is rather artificial,
and both can be employed.

The coordination numbers for metal atoms in a molecule or complex are
commonly 4, 5, and 6, but all values from 2 to 9 are known and a few
examples of higher coordination numbers have been reported. In contrast,
common coordination numbers in the solid state are 3, 4, 6, 8, and 12. For
example, the atom in the center of body-centered cubic lattice has a
coordination number of 8, because it touches the eight atoms at the corners
of the unit cell, while an atom in a simple cubic structure would have a
coordination number of 6. In both fcc and hcp lattices each of the atoms
have a coordination number of 12.

Octahedral and tetrahedral vacancies

As was mentioned above, the packing fraction in both fcc and hcp cells is
74.05%, leaving 25.95% of the volume unfilled. The unfilled lattice sites
(interstices) between the atoms in a cell are called interstitial sites or
vacancies. The shape and relative size of these sites is important in
controlling the position of additional atoms. In both fcc and hcp cells most
of the space within these atoms lies within two different sites known as
octahedral sites and tetrahedral sites. The difference between the two lies in
their “coordination number”, or the number of atoms surrounding each site.
Tetrahedral sites (vacancies) are surrounded by four atoms arranged at the
comers of a tetrahedron. Similarly, octahedral sites are surrounded by six
atoms which make-up the apices of an octahedron. For a given close packed
lattice an octahedral vacancy will be larger than a tetrahedral vacancy.

Within a face centered cubic lattice, the eight tetrahedral sites are positioned
within the cell, at the general fractional coordinate of (°/4,"/4,"/4) where n =
1 or 3, e.g., (7/4, /4,!/4), (/4,"/4,7/4), etc. The octahedral sites are located at
the center of the unit cell (1/5,"/5,'/5), as well as at each of the edges of the
cell, e.g., (4/5,0,0). In the hexagonal close packed system, the tetrahedral
sites are at (0,0,°/g) and (1/3,/3,’/g), and the octahedral sites are at
(1/3,"/3,'/4) and all symmetry equivalent positions.

Important structure types

The majority of crystalline materials do not have a structure that fits into the
one atom per site simple Bravais lattice. A number of other important
crystal structures are found, however, only a few of these crystal structures
are those of which occur for the elemental and compound semiconductors
and the majority of these are derived from fcc or hcp lattices. Each
structural type is generally defined by an archetype, a material (often a
naturally occurring mineral) which has the structure in question and to
which all the similar materials are related. With regard to commonly used
elemental and compound semiconductors the important structures are
diamond, zinc blende, Wurtzite, and to a lesser extent chalcopyrite.
However, rock salt, B-tin, cinnabar and cesium chloride are observed as
high pressure or high temperature phases and are therefore also discussed.

The following provides a summary of these structures. Details of the full
range of solid-state structures are given elsewhere.

Diamond Cubic

The diamond cubic structure consists of two interpenetrating face-centered
cubic lattices, with one offset '/, of a cube along the cube diagonal. It may
also be described as face centered cubic lattice in which half of the
tetrahedral sites are filled while all the octahedral sites remain vacant. The
diamond cubic unit cell is shown in [link]. Each of the atoms (e.g., C) is
four coordinate, and the shortest interatomic distance (C-C) may be
determined from the unit cell parameter (a).

Equation:

Unit cell structure of a
diamond cubic lattice
showing the two
interpenetrating face-
centered cubic lattices.

Zinc blende

This is a binary phase (ME) and is named after its archetype, a common
mineral form of zinc sulfide (ZnS). As with the diamond lattice, zinc blende
consists of the two interpenetrating fcc lattices. However, in zinc blende one
lattice consists of one of the types of atoms (Zn in ZnS), and the other
lattice is of the second type of atom (S in ZnS). It may also be described as
face centered cubic lattice of S atoms in which half of the tetrahedral sites
are filled with Zn atoms. All the atoms in a zinc blende structure are 4-
coordinate. The zinc blende unit cell is shown in [link]. A number of inter-
atomic distances may be calculated for any material with a zinc blende unit
cell using the lattice parameter (a).

Equation:

Zn-S = av3 = 0.422a
4

Equation:

Zn-Zn = S-S = a_= 0.707 a
v2

Unit cell structure of a

zinc blende (ZnS) lattice.
Zinc atoms are shown in
green (small), sulfur
atoms shown in red
(large), and the dashed
lines show the unit cell.

Chalcopyrite

The mineral chalcopyrite CuFeS, is the archetype of this structure. The
structure is tetragonal (a = b#c, a= 8 = y = 90°, and is essentially a
superlattice on that of zinc blende. Thus, is easiest to imagine that the
chalcopyrite lattice is made-up of a lattice of sulfur atoms in which the
tetrahedral sites are filled in layers, ...FeCuCuFe..., etc. ({link]). In such an
idealized structure c = 2a, however, this is not true of all materials with
chalcopyrite structures.

Unit cell structure of a
chalcopyrite lattice.
Copper atoms are shown
in blue, iron atoms are
shown in green and sulfur
atoms are shown in
yellow. The dashed lines
show the unit cell.

Rock salt

As its name implies the archetypal rock salt structure is NaCl (table salt). In
common with the zinc blende structure, rock salt consists of two
interpenetrating face-centered cubic lattices. However, the second lattice is
offset 1/2a along the unit cell axis. It may also be described as face centered
cubic lattice in which all of the octahedral sites are filled, while all the
tetrahedral sites remain vacant, and thus each of the atoms in the rock salt

structure are 6-coordinate. The rock salt unit cell is shown in [link]. A
number of inter-atomic distances may be calculated for any material with a
rock salt structure using the lattice parameter (a).

Equation:

Na-Cl = a = 05a
2

Equation:

Na-Na = CI-Cl = a = 0.707a
v2

Unit cell structure of a
rock salt lattice. Sodium
ions are shown in purple

(small spheres) and
chloride ions are shown
in red (large spheres).

Cinnabar

Cinnabar, named after the archetype mercury sulfide, Hg§S, is a distorted
rock salt structure in which the resulting cell is rhombohedral (trigonal)
with each atom having a coordination number of six.

Wurtzite

This is a hexagonal form of the zinc sulfide. It is identical in the number of
and types of atoms, but it is built from two interpenetrating hcp lattices as
opposed to the fcc lattices in zinc blende. As with zinc blende all the atoms
in a wurtzite structure are 4-coordinate. The wurtzite unit cell is shown in
[link]. A number of inter atomic distances may be calculated for any
material with a wurtzite cell using the lattice parameter (a).

Equation:

Zn-S = av3/8 = 0.612a = 3c = 0375c¢
8

Equation:

Zn-Zn = S-S = a = 1.632c

However, it should be noted that these formulae do not necessarily apply
when the ratio a/c is different from the ideal value of 1.632.

Unit cell structure of a
wurtzite lattice. Zinc
atoms are shown in green
(small spheres), sulfur
atoms shown in red (large
spheres), and the dashed
lines show the unit cell.

Cesium Chloride

The cesium chloride structure is found in materials with large cations and
relatively small anions. It has a simple (primitive) cubic cell ([link]) with a
chloride ion at the corners of the cube and the cesium ion at the body center.
The coordination numbers of both Cs* and Cl’, with the inner atomic
distances determined from the cell lattice constant (a).

Equation:

Cs-Cl = ayv3 = 0.8664
2

Equation:

Cs-Cs = CI-Cl =a

B-Tin.

The room temperature allotrope of tin is B-tin or white tin. It has a
tetragonal structure, in which each tin atom has four nearest neighbors (Sn-
Sn = 3.016 A) arranged in a very flattened tetrahedron, and two next nearest
neighbors (Sn-Sn = 3.175 A). The overall structure of B-tin consists of
fused hexagons, each being linked to its neighbor via a four-membered Sn,
ring.

Defects in crystalline solids

Up to this point we have only been concerned with ideal structures for
crystalline solids in which each atom occupies a designated point in the
crystal lattice. Unfortunately, defects ordinarily exist in equilibrium
between the crystal lattice and its environment. These defects are of two
general types: point defects and extended defects. As their names imply,
point defects are associated with a single crystal lattice site, while extended
defects occur over a greater range.

Point defects: “too many or too few” or “just plain wrong”

Point defects have a significant effect on the properties of a semiconductor,
so it is important to understand the classes of point defects and the
characteristics of each type. [link] summarizes various classes of native
point defects, however, they may be divided into two general classes;
defects with the wrong number of atoms (deficiency or surplus) and defects
where the identity of the atoms is incorrect.

:
-

(a) perfect lattice (b) interstitial impurity

:

(c) cation vacancy (d) anion vacancy

(e) substitution of cation (f) substitution of anion

(g) Ba antisite defect (h) Ag antisite defect

Point defects in a crystal lattice.

Interstitial Impurity

An interstitial impurity occurs when an extra atom is positioned in a lattice
site that should be vacant in an ideal structure ([{link]b). Since all the
adjacent lattice sites are filled the additional atom will have to squeeze itself
into the interstitial site, resulting in distortion of the lattice and alteration in
the local electronic behavior of the structure. Small atoms, such as carbon,

will prefer to occupy these interstitial sites. Interstitial impurities readily
diffuse through the lattice via interstitial diffusion, which can result in a
change of the properties of a material as a function of time. Oxygen
impurities in silicon generally are located as interstitials.

Vacancies

The converse of an interstitial impurity is when there are not enough atoms
in a particular area of the lattice. These are called vacancies. Vacancies exist
in any material above absolute zero and increase in concentration with
temperature. In the case of compound semiconductors, vacancies can be
either cation vacancies ({link]c) or anion vacancies ([{link]d), depending on
what type of atom are “missing”.

Substitution

Substitution of various atoms into the normal lattice structure is common,
and used to change the electronic properties of both compound and
elemental semiconductors. Any impurity element that is incorporated
during crystal growth can occupy a lattice site. Depending on the impurity,
substitution defects can greatly distort the lattice and/or alter the electronic
structure. In general, cations will try to occupy cation lattice sites ([link]e),
and anion will occupy the anion site ({link]f). For example, a zinc impurity
in GaAs will occupy a gallium site, if possible, while a sulfur, selenium and
tellurium atoms would all try to substitute for an arsenic. Some impurities
will occupy either site indiscriminately, e.g., Si and Sn occupy both Ga and
As sites in GaAs.

Antisite Defects

Antisite defects are a particular form of substitution defect, and are unique
to compound semiconductors. An antisite defect occurs when a cation is
misplaced on an anion lattice site or vice versa ([link]g and h). Dependant

on the arrangement these are designated as either Ap antisite defects or Ba
antisite defects. For example, if an arsenic atom is on a gallium lattice site
the defect would be an Asc, defect. Antisite defects involve fitting into a
lattice site atoms of a different size than the rest of the lattice, and therefore
this often results in a localized distortion of the lattice. In addition, cations
and anions will have a different number of electrons in their valence shells,
so this substitution will alter the local electron concentration and the
electronic properties of this area of the semiconductor.

Extended Defects: Dislocations in a Crystal Lattice

Extended defects may be created either during crystal growth or as a
consequence of stress in the crystal lattice. The plastic deformation of
crystalline solids does not occur such that all bonds along a plane are
broken and reformed simultaneously. Instead, the deformation occurs
through a dislocation in the crystal lattice. [link] shows a schematic
representation of a dislocation in a crystal lattice. Two features of this type
of dislocation are the presence of an extra crystal plane, and a large void at
the dislocation core. Impurities tend to segregate to the dislocation core in
order to relieve strain from their presence.

extra net plane
0 . direction of slip
—_____ >»

dislocation
core

Dislocation in a crystal lattice.

Epitaxy

Epitaxy, is a transliteration of two Greek words epi, meaning "upon", and
taxis, meaning "ordered". With respect to crystal growth it applies to the
process of growing thin crystalline layers on a crystal substrate. In epitaxial
growth, there is a precise crystal orientation of the film in relation to the
substrate. The growth of epitaxial films can be done by a number of
methods including molecular beam epitaxy, atomic layer epitaxy, and
chemical vapor deposition, all of which will be described later.

Epitaxy of the same material, such as a gallium arsenide film on a gallium
arsenide substrate, is called homoepitaxy, while epitaxy where the film and
substrate material are different is called heteroepitaxy. Clearly, in
homoepitaxy, the substrate and film will have the identical structure,
however, in heteroepitaxy, it is important to employ where possible a
substrate with the same structure and similar lattice parameters. For
example, zinc selenide (zinc blende, a = 5.668 A) is readily grown on
gallium arsenide (zinc blende, a = 5.653 A). Alternatively, epitaxial crystal
growth can occur where there exists a simple relationship between the
structures of the substrate and crystal layer, such as is observed between
AlyO3 (100) on Si (100). Whichever route is chosen a close match in the
lattice parameters is required, otherwise, the strains induced by the lattice
mismatch results in distortion of the film and formation of dislocations. If
the mismatch is significant epitaxial growth is not energetically favorable,
causing a textured film or polycrystalline untextured film to be grown. As a
general rule of thumb, epitaxy can be achieved if the lattice parameters of
the two materials are within about 5% of each other. For good quality
epitaxy, this should be less than 1%. The larger the mismatch, the larger the
strain in the film. As the film gets thicker and thicker, it will try to relieve
the strain in the film, which could include the loss of epitaxy of the growth
of dislocations. It is important to note that the <100> directions of a film
must be parallel to the <100> direction of the substrate. In some cases, such
as Fe on MgO, the [111] direction is parallel to the substrate [100]. The
epitaxial relationship is specified by giving first the plane in the film that is
parallel to the substrate [100].

Bibliography

e International Tables for X-ray Crystallography. Vol. IV; Kynoch
Press: Birmingham, UK (1974).

¢ B. F. G. Johnson, in Comprehensive Inorganic Chemistry, Pergamon
Press, Vol. 4, Chapter 52 (1973).

e A. R. West, Solid State Chemistry and its Applications, Wiley, New
York (1984).

Structures of Element and Compound Semiconductors

Introduction

A single crystal of either an elemental (e.g., silicon) or compound (e.g.,
gallium arsenide) semiconductor forms the basis of almost all
semiconductor devices. The ability to control the electronic and opto-
electronic properties of these materials is based on an understanding of their
structure. In addition, the metals and many of the insulators employed
within a microelectronic device are also crystalline.

Group IV (14) elements

Each of the semiconducting phases of the group IV (14) elements, C
(diamond), Si, Ge, and a-Sn, adopt the diamond cubic structure ((link]).
Their lattice constants (a, A) and densities (p, g/cm?) are given in [link].

Unit cell structure of a
diamond cubic lattice
showing the two
interpenetrating face-
centered cubic lattices.

Lattice parameter, a

Element Density (g/cm?
carbon

(arsond) 3.56683(1) 3.51525

silicon 5.4310201(3) 2.319002
germanium 5.657906(1) 5.3234

tin (a-Sn) 6.4892(1) 7.285

Lattice parameters and densities (measured at 298 K) for the diamond cubic
forms of the group IV (14) elements.

As would be expected the lattice parameter increase in the order C < Si <
Ge < a-Sn. Silicon and germanium form a continuous series of solid
solutions with gradually varying parameters. It is worth noting the high
degree of accuracy that the lattice parameters are known for high purity
crystals of these elements. In addition, it is important to note the
temperature at which structural measurements are made, since the lattice
parameters are temperature dependent ([link]). The lattice constant (a), in
A, for high purity silicon may be calculated for any temperature (T) over
the temperature range 293 - 1073 K by the formula shown below.

ay = 5.4304 + 1.8138 X 10° (T - 298.15 K) + 1.542 X 10°9 (T — 298.15 K)

(a) 5.447
5.444

5.44]

aA) 5 a3
5.435

5.432

5.429
0 100 200 300 400 500 600 700 800

Temperature (°C)

(b) 5.69

5.66

0 100 200 300 400 500 600 700 800
Temperature (°C)

Temperature dependence of the
lattice parameter for (a) Si and
(b) Ge.

Even though the diamond cubic forms of Si and Ge are the only forms of
direct interest to semiconductor devices, each exists in numerous crystalline
high pressure and meta-stable forms. These are described along with their
interconversions, in [link].

Phase

Sil

Si II

Si III

Si TV

Si V

Si VI

Ge I

Ge II

Ge III

Ge IV

Structure
diamond cubic

grey tin
structure

cubic

hexagonal

unidentified

hexagonal
close packed

diamond cubic

B-tin structure

tetragonal

body centered
cubic

Remarks
stable at normal pressure

formed from Si I or Si V above 14
GPa

metastable, formed from Si II above
10 GPa

stable above 34 GPa, formed from Si
II above 16 GPa

stable above 45 GPa

low-pressure phase
formed from Ge I above 10 GPa

formed by quenching Ge II at low
pressure

formed by quenching Ge II to 1 atm
at 200 K

High pressure and metastable phases of silicon and germanium.

Group ITI-V (13-15) compounds

The stable phases for the arsenides, phosphides and antimonides of
aluminum, gallium and indium all exhibit zinc blende structures ([link]). In
contrast, the nitrides are found as wurtzite structures (e.g., [link]). The
structure, lattice parameters, and densities of the III-V compounds are given

in [link]. It is worth noting that contrary to expectation the lattice parameter
of the gallium compounds is smaller than their aluminum homolog; for
GaAs a = 5.653 A; AlAs a = 5.660 A. As with the group IV elements the
lattice parameters are highly temperature dependent; however, additional
variation arises from any deviation from absolute stoichiometry. These
effects are shown in [link].

Unit cell structure of a
zinc blende (ZnS) lattice.
Zinc atoms are shown in

green (small), sulfur
atoms shown in red

(large), and the dashed
lines show the unit cell.

Unit cell structure of a
wurtzite lattice. Zinc
atoms are shown in green
(small), sulfur atoms
shown in red (large), and
the dashed lines show the

unit cell.
Compound Structure
AIN wurtzite
zinc
og blende
AlAs zinc

blende

Lattice
parameter (A)

a = 3.11(1), c=
4.98(1)

a = 5.4635(4)
a = 5.660

Density
(g/cm?)

3.200

2.40(1)

3.760

AlSb zinc a = 6.1355(1) 4.26

blende
GaN wurtzite ae Ee
GaP ae - a = 5.4505(2) 4.138
GaAs ae a a = 5.65325(2) 5.3176(3)
InN wurtzite ae oi 6.81
InP a a a = 5.868(1) 4.81
InAs he ” a = 6.0583 5.667
InSb ae 7 a = 6.47937 5.7747(4)

Lattice parameters and densities (measured at 298 K) for the II-V (13-15)
compound semiconductors. Estimated standard deviations given in
parentheses.

stoichiometric

a(A) 5.

0 10 20 30 8 40 50 60 70
Temperature (°C)

Temperature dependence of the lattice
parameter for stoichiometric GaAs
and crystals with either Ga or As
excess.

The homogeneity of structures of alloys for a wide range of solid solutions
to be formed between ITI-V compounds in almost any combination. Two
classes of ternary alloys are formed: III,-II,_,-V (e.g., Al,-Ga,.,-As) and
IT-V1_,-Vx (e.g., Ga-Asj_,-P,) . While quaternary alloys of the type III,-
IIT,_,-V,-V1-y allow for the growth of materials with similar lattice
parameters, but a broad range of band gaps. A very important ternary alloy,
especially in optoelectronic applications, is Al,-Ga,_,-As and its lattice
parameter (a) is directly related to the composition (x).

d = 5.6533 + 0.0078 x

Not all of the III-V compounds have well characterized high-pressure
phases. however, in each case where a high-pressure phase is observed the
coordination number of both the group III and group V element increases
from four to six. Thus, AIP undergoes a zinc blende to rock salt
transformation at high pressure above 170 kbar, while AlSb and GaAs form
orthorhombic distorted rock salt structures above 77 and 172 kbar,

respectively. An orthorhombic structure is proposed for the high-pressure
form of InP (>133 kbar). Indium arsenide (InAs) undergoes two-phase
transformations. The zinc blende structure is converted to a rock salt
structure above 77 kbar, which in turn forms a B-tin structure above 170
kbar.

Group II-VI (12-16) compounds

The structures of the II-VI compound semiconductors are less predictable
than those of the III-V compounds (above), and while zinc blende structure
exists for almost all of the compounds there is a stronger tendency towards
the hexagonal wurtzite form. In several cases the zinc blende structure is
observed under ambient conditions, but may be converted to the wurtzite
form upon heating. In general the wurtzite form predominates with the
smaller anions (e.g., oxides), while the zinc blende becomes the more stable
phase for the larger anions (e.g., tellurides). One exception is mercury
sulfide (HgS) that is the archetype for the trigonal cinnabar phase. [link]
lists the stable phase of the chalcogenides of zinc, cadmium and mercury,
along with their high temperature phases where applicable. Solid solutions
of the II-VI compounds are not as easily formed as for the III-V
compounds; however, two important examples are ZnS,Se,_, and Cd,Hg,.
le:

Lattice Density
Compound Structure parameter (A) (g/cm?)
ZINC =
7nS Tae a=5.410 4.075
wurtzite ae 087

6.260

ZnSe Zinc a = 5.668 oa

blende
Zinc -
ZntTe eiende a= 6.10 5.636
: a = 4.136, c =
CdS wurtzite 6.714 4.82
. a = 4.300, c =
CdSe wurtzite 7011 5.81
Zinc _
CdTe Pleada a = 6.482 5.87
; a=4.149,c=
Hgs cinnabar 9.495
ane - a= 5.851 7.73
Zinc _
HgSe hewis a = 6.085 8.25
Zinc _
HgTe beade a = 6.46 8.07

Lattice parameters and densities (measured at 298 K) for the II-VI (12-16)
compound semiconductors.

The zinc chalcogenides all transform to a cesium chloride structure under
high pressures, while the cadmium compounds all form rock salt high-
pressure phases ([link]). Mercury selenide (HgSe) and mercury telluride
(HgTe) convert to the mercury sulfide archetype structure, cinnabar, at high
pressure.

Unit cell structure of a
rock salt lattice. Sodium
ions are shown in purple

and chloride ions are
shown in red.

I-III-VI, (11-13-16) compounds

Nearly all I-III-VI, compounds at room temperature adopt the chalcopyrite
structure ([link]). The cell constants and densities are given in [link].
Although there are few reports of high temperature or high-pressure phases,
AgInS> has been shown to exist as a high temperature orthorhombic
polymorph (a = 6.954, b = 8.264, and c = 6.683 A), and AgInTe, forms a
cubic phase at high pressures.

Unit cell structure of a
chalcopyrite lattice.
Copper atoms are shown
in blue, iron atoms are
shown in green and sulfur
atoms are shown in
yellow. The dashed lines
show the unit cell.

Lattice Lattice
Compound parameter a parameter c

(A) (A) (g.cm’)

Density

CuAlS> 5.02 10.430 3.45

CuAlSep 9.61 10.92 4.69
CuAlTe, 5.96 177 9.47
CuGaS» 9.39 10.46 4.38
CuGaSe» 9.61 11.00 rel
CuGatTep 6.00 11.93 9.95
CulnS> Di02 11.08 4.74
CulnSe, 5.78 11.55 Deld-
CulnTe, 6.17 12.34 6.10
AgAIS» 6.30 11.84 6.15
AgGaS» DLO 10.29 4.70
AgGaSe> 5.98 10.88 5.70
AgGatTeo 6.29 11.95 6.08
AgInS» 5.82 11.17 4.97
AgInSe> 6.095 11.69 5.82
AginTe> 6.43 12,09 6.96

Chalcopyrite lattice parameters and densities (measured at 298 K) for the I-
II-VI compound semiconductors. Lattice parameters for tetragonal cell.

Of the I-ITI-VI, compounds, the copper indium chalcogenides (CuInE>) are
certainly the most studied for their application in solar cells. One of the

advantages of the copper indium chalcogenide compounds is the formation
of solid solutions (alloys) of the formula CulnE>_,E',, where the
composition variable (x) varies from 0 to 2. The CulnS5_,Se, and CulnSe >.
x le, systems have also been examined, as has the CuGa,Inj.yS7_,Sex
quaternary system. As would be expected from a consideration of the
relative ionic radii of the chalcogenides the lattice parameters of the
CulnS>_,Se, alloy should increase with increased selenium content.
Vergard's law requires the lattice constant for a linear solution of two
semiconductors to vary linearly with composition (e.g., as is observed for
Al,Ga;_,As), however, the variation of the tetragonal lattice constants (a
and c) with composition for CulnS>_,S, are best described by the parabolic
relationships.

a = 5.532 + 0.0801 x + 0.0260 x?

c = 11.156 + 0.1204 x + 0.0611 x?

A similar relationship is observed for the CulnSe_,Te, alloys.
a = 5.783 + 0.1560 x + 0.0212 x?

c = 11.628 + 0.3340 x + 0.0277 x?

The large difference in ionic radii between S and Te (0.37 A) prevents
formation of solid solutions in the CulnS»_,Te, system, however, the single
alloy CulnS, 5Teg 5 has been reported.

Orientation effects

Once single crystals of high purity silicon or gallium arsenide are produced
they are cut into wafers such that the exposed face of these wafers is either
the crystallographic {100} or {111} planes. The relative structure of these
surfaces are important with respect to oxidation, etching and thin film
growth. These processes are orientation-sensitive; that is, they depend on
the direction in which the crystal slice is cut.

Atom density and dangling bonds

The principle planes in a crystal may be differentiated in a number of ways,
however, the atom and/or bond density are useful in predicting much of the
chemistry of semiconductor surfaces. Since both silicon and gallium
arsenide are fcc structures and the {100} and {111} are the only
technologically relevant surfaces, discussions will be limited to fcc {100}
and {111}.

The atom density of a surface may be defined as the number of atoms per
unit area. [link] shows a schematic view of the {111} and {100} planes in a
fcc lattice. The {111} plane consists of a hexagonal close packed array in
which the crystal directions within the plane are oriented at 60° to each
other. The hexagonal packing and the orientation of the crystal directions
are indicated in [link]b as an overlaid hexagon. Given the intra-planar inter-
atomic distance may be defined as a function of the lattice parameter, the
area of this hexagon may be readily calculated. For example in the case of
silicon, the hexagon has an area of 38.30 A*. The number of atoms within
the hexagon is three: the atom in the center plus 1/3 of each of the six atoms
at the vertices of the hexagon (each of the atoms at the hexagons vertices is
shared by three other adjacent hexagons). Thus, the atom density of the
{111} plane is calculated to be 0.0783 A’. Similarly, the atom density of
the {100} plane may be calculated. The {100} plane consists of a square
array in which the crystal directions within the plane are oriented at 90° to
each other. Since the square is coincident with one of the faces of the unit
cell the area of the square may be readily calculated. For example in the
case of silicon, the square has an area of 29.49 A*. The number of atoms
within the square is 2: the atom in the center plus 1/4 of each of the four
atoms at the vertices of the square (each of the atoms at the corners of the
square are shared by four other adjacent squares). Thus, the atom density of
the {100} plane is calculated to be 0.0678 A-*. While these values for the
atom density are specific for silicon, their ratio is constant for all diamond
cubic and zinc blende structures: {100}:{111} = 1:1.155. In general, the
fewer dangling bonds the more stable a surface structure.

Schematic representation of the (111) and
(100) faces of a face centered cubic (fcc)
lattice showing the relationship between the
close packed rows.

An atom inside a crystal of any material will have a coordination number
(n) determined by the structure of the material. For example, all atoms
within the bulk of a silicon crystal will be in a tetrahedral four-coordinate
environment (n = 4). However, at the surface of a crystal the atoms will not
make their full compliment of bonds. Each atom will therefore have less
nearest neighbors than an atom within the bulk of the material. The missing
bonds are commonly called dangling bonds. While this description is not
particularly accurate it is, however, widely employed and as such will be
used herein. The number of dangling bonds may be defined as the
difference between the ideal coordination number (determined by the bulk
crystal structure) and the actual coordination number as observed at the
surface.

[link] shows a section of the {111} surfaces of a diamond cubic lattice
viewed perpendicular to the {111} plane. The atoms within the bulk have a
coordination number of four. In contrast, the atoms at the surface (e.g., the
atom shown in blue in [link]) are each bonded to just three other atoms (the
atoms shown in red in [link]), thus each surface atom has one dangling
bond. As can be seen from [link], which shows the atoms at the {100}
surface viewed perpendicular to the {100} plane, each atom at the surface
(e.g., the atom shown in blue in [link]) is only coordinated to two other
atoms (the atoms shown in red in [link]), leaving two dangling bonds per

atom. It should be noted that the same number of dangling bonds are found
for the {111} and {100} planes of a zinc blende lattice. The ratio of
dangling bonds for the {100} and {111} planes of all diamond cubic and
zinc blende structures is {100}:{111} = 2:1. Furthermore, since the atom
densities of each plane are known then the ratio of the dangling bond
densities is determined to be: {100}:{111} = 1:0.577.

A section of the {111}
surfaces of a diamond
cubic lattice viewed
perpendicular to the
{111} plane.

A section of the {100}
surface of a diamond
cubic lattice viewed
perpendicular to the

{100} plane.

Silicon

For silicon, the {111} planes are closer packed than the {100} planes. As a
result, growth of a silicon crystal is therefore slowest in the <111>
direction, since it requires laying down a close packed atomic layer upon
another layer in its closest packed form. As a consequence <111> Si is the
easiest to grow, and therefore the least expensive.

The dissolution or etching of a crystal is related to the number of broken
bonds already present at the surface: the fewer bonds to be broken in order
to remove an individual atom from a crystal, the easier it will be to dissolve
the crystal. As a consequence of having only one dangling bond (requiring
three bonds to be broken) etching silicon is slowest in the <111> direction.
The electronic properties of a silicon wafer are also related to the number of
dangling bonds.

Silicon microcircuits are generally formed on a single crystal wafer that is
diced after fabrication by either sawing part way through the wafer
thickness or scoring (scribing) the surface, and then physically breaking.
The physical breakage of the wafer occurs along the natural cleavage
planes, which in the case of silicon are the {111} planes.

Gallium arsenide

The zinc blende lattice observed for gallium arsenide results in additional
considerations over that of silicon. Although the {100} plane of GaAs is
structurally similar to that of silicon, two possibilities exist: a face
consisting of either all gallium atoms or all arsenic atoms. In either case the
surface atoms have two dangling bonds, and the properties of the face are
independent of whether the face is gallium or arsenic.

The {111} plane also has the possibility of consisting of all gallium or all
arsenic. However, unlike the {100} planes there is a significant difference
between the two possibilities. [link] shows the gallium arsenide structure
represented by two interpenetrating fcc lattices. The [111] axis is vertical
within the plane of the page. Although the structure consists of alternate
layers of gallium and arsenic stacked along the [111] axis, the distance
between the successive layers alternates between large and small. Assigning
arsenic as the parent lattice the order of the layers in the [111] direction is

As— Ga-As— Ga-As~— Ga, while in the | 111 | direction the layers are

ordered, Ga-As-Ga— As-Ga— As ({link]). In silicon these two directions are
of course identical. The surface of a crystal would be either arsenic, with
three dangling bonds, or gallium, with one dangling bond. Clearly, the latter
is energetically more favorable. Thus, the (111) plane shown in [link] is

called the (111) Ga face. Conversely, the fii plane would be either

gallium, with three dangling bonds, or arsenic, with one dangling bond.

Again, the latter is energetically more favorable and the fii plane is

therefore called the (111) As face.

The (111) Ga face of
GaAs showing a surface
layer containing gallium

atoms (green) with one
dangling bond per
gallium and three bonds
to the arsenic atoms (red)
in the lower layer.

The (111) As is distinct from that of (111) Ga due to the difference in the
number of electrons at the surface. As a consequence, the (111) As face
etches more rapidly than the (111) Ga face. In addition, surface evaporation
below 770 °C occurs more rapidly at the (111) As face.

Bibliography

e M. Baublitz and A. L. Ruoff, J. Appl. Phys., 1982, 53, 6179.

J. C. Jamieson, Science, 1963, 139, 845.

C. C. Landry, J. Lockwood, and A. R. Barron, Chem. Mater., 1995, 7,
699.

e M. Robbins, J. C. Phillips, and V. G. Lambrecht, J. Phys. Chem.
Solids, 1973, 34, 1205.

D. Sridevi and K. V. Reddy, Mat. Res. Bull., 1985, 20, 929.

Y. K. Vohra, S. T. Weir, and A. L. Ruoff, Phys. Rev. B, 1985, 31, 7344.
e W. M. Yin and R. J. Paff, J. Appl. Phys., 1973, 45, 1456.

Introduction to Bipolar Transistors

Note:This module is adapted from the Connexions module entitled
Introduction to Bipolar Transistors by Bill Wilson.

Let's leave the world of two terminal devices (which are all called diodes by
the way; diode just means two-terminals) and venture into the much more
interesting world of three terminals. The first device we will look at is
called the bipolar transistor. Consider the structure shown in [link]:

+
Emitter | Base | Collector oe
n+ p n

Structure of a npn bipolar
transistor.

The device consists of three layers of silicon, a heavily doped n-type layer
called the emitter, a moderately doped p-type layer called the base, and
third, more lightly doped layer called the collector. In a biasing (applied DC
potential) configuration called forward active biasing, the emitter-base
junction is forward biased, and the base-collector junction is reverse biased.
[link] shows the biasing conventions we will use. Both bias voltages are
referenced to the base terminal. Since the base-emitter junction is forward
biased, and since the base is made of p-type material, Vp must be negative.
On the other hand, in order to reverse bias the base-collector junction Vcp
will be a positive voltage.

Forward active biasing of a npn
bipolar transistor.

Now, let's draw the band-diagram for this device. At first this might seem
hard to do, but we know what forward and reverse biased band diagrams
look like, so we'll just stick one of each together. We show this in [link],
which is a very busy figure, but it is also very important, because it shows
all of the important features in the operation the transistor. Since the base-
emitter junction is forward biased, electrons will go from the (n-type)
emitter into the base. Likewise, some holes from the base will be injected
into the emitter.

‘Ee

Band diagram and carrier fluxes
in a bipolar transistor.

In [link], we have two different kinds of arrows. The open arrows which are
attached to the carriers, show us which way the carrier is moving. The solid
arrows which are labeled with some kind of subscripted J, represent current
flow. We need to do this because for holes, motion and current flow are in
the same direction, while for electrons, carrier motion and current flow are
in opposite directions.

Just as we saw in the last chapter, the electrons which are injected into the
base diffuse away from the emitter-base junction towards the (reverse
biased) base-collector junction. As they move through the base, some of the
electrons encounter holes and recombine with them. Those electrons which
do get to the base-collector junction run into a large electric field which
sweeps them out of the base and into the collector. They "fall" down the
large potential drop at the junction.

These effects are all seen in [link], with arrows representing the various
currents which are associated with each of the carriers fluxes. I, represents
the current associated with the electron injection into the base, i.e., it points
in the opposite direction from the motion of the electrons, since electrons
have a negative charge. Iz, represents the current associated with holes
injection into the emitter from the base. Ip, represents recombination
current in the base, while Jc, represents the electron current going into the
collector. It should be easy for you to see that:

Equation:
Ip = Ige + Len
Equation:
Ip = Ign + Lpr
Equation:
Ic = Ice

In a "good" transistor, almost all of the current across the base-emitter
junction consists of electrons being injected into the base. The transistor
engineer works hard to design the device so that very little emitter current is

made up of holes coming from the base into the emitter. The transistor is
also designed so that almost all of those electrons which are injected into
the base make it across to the base-collector reverse-biased junction. Some
recombination is unavoidable, but things are arranged so as to minimize this
effect.

Basic MOS Structure

Note:This module is adapted from the Connexions module entitled Basic
MOS Structure by Bill Wilson.

[link] shows the basic steps necessary to make the MOS structure. It will
help us in our understanding if we now rotate our picture so that it is
pointing sideways in our next few drawings. [link] shows the rotated
structure. Note that in the p-silicon we have positively charged mobile
holes, and negatively charged, fixed acceptors. Because we will need it
later, we have also shown the band diagram for the semiconductor below
the sketch of the device. Note that since the substrate is p-type, the Fermi

level is located down close to the valance band.
polysilicon

SiO» Op + heat ~ SiH4 + heat
—— PELL EL ELIE ELD

Formation of the metal-oxide-
semiconductor (MOS) structure.

Basic metal-oxide-
semiconductor
(MOS) structure.

Let us now place a potential between the gate and the silicon substrate.
Suppose we make the gate negative with respect to the substrate. Since the
substrate is p-type, it has a lot of mobile, positively charged holes in it.
Some of them will be attracted to the negative charge on the gate, and move
over to the surface of the substrate. This is also reflected in the band
diagram shown in [link]. Remember that the density of holes is
exponentially proportional to how close the Fermi level is to the valence
band edge. We see that the band diagram has been bent up slightly near the
surface to reflect the extra holes which have accumulated there.

Applying a
negative gate
voltage to a basic
metal-oxide-
semiconductor
(MOS) structure.

An electric field will develop between the positive holes and the negative
gate charge. Note that the gate and the substrate form a kind of parallel
plate capacitor, with the oxide acting as the insulating layer in-between
them. The oxide is quite thin compared to the area of the device, and so it is
quite appropriate to assume that the electric field inside the oxide is a
uniform one. (We will ignore fringing at the edges.) The integral of the
electric field is just the applied gate voltage V,. If the oxide has a thickness
Xox then since E,, is uniform, it is given by, [link].

Equation:

Vo

Lox

Fox =

If we focus in on a small part of the gate, we can make a little "pill" box
which extends from somewhere in the oxide, across the oxide/gate interface
and ends up inside the gate material someplace. The pill-box will have an
area As. Now we will invoke Gauss' law which we reviewed earlier. Gauss'
law simply says that the surface integral over a closed surface of the
displacement vector D (which is, of course, € x E) is equal to the total
charge enclosed by that surface. We will assume that there is a surface
charge density -Q, Coulombs/cm? on the surface of the gate electrode
({link]). The integral form of Gauss' Law is just:

Equation:

f exk dS= Qencl

surface charge
density Qg

Electric Field

‘pill box"
with area 4s

Finding the surface
charge density.

Note that we have used €,,F in place of D. In this particular set-up the
integral is easy to perform, since the electric field is uniform, and only
pointing in through one surface - it terminates on the negative surface
charge inside the pill-box. The charge enclosed in the pill box is just -
(QgAs), and so we have (keeping in mind that the surface integral of a
vector pointing into the surface is negative), [link], or [link].
Equation:

fey E AS = —(€oxEoxA(s))
— (Q,A(s))
Equation:
EoxEox = Qs

Now, we can use [link] to get [link] or [link].

Equation:
EoxVg Q
Lox
Equation:
Qs Eox __
Tr = “ox
Vy Box

The quantity c,, is called the oxide capacitance. It has units of Farads/cm?,
so it is really a capacitance per unit area of the oxide. The dielectric
constant of silicon dioxide, €,,, is about 3.3 x 10°!8 F/cm. A typical oxide
thickness might be 250 A (or 2.5 x 10° cm). In this case, c,, would be
about 1.30 x 10°’ F/cm?. The units we are using here, while they might

seem a little arbitrary and confusing, are the ones most commonly used in
the semiconductor business.

The most useful form of [link] is when it is turned around, [link], as it gives
us a way to find the charge on the gate in terms of the gate potential. We
will use this equation later in our development of how the MOS transistor
really works.

Equation:

Qs = Con

It turns out we have not done anything very useful by apply a negative
voltage to the gate. We have drawn more holes there in what is called an
accumulation layer, but that is not helping us in our effort to create a layer
of electrons in the MOSFET which could electrically connect the two n-
regions together.

Let's turn the battery around and apply a positive voltage to the gate
([link]). Actually, let's take the battery out for now, and just let V, be a
positive value, relative to the substrate which will tie to ground. Making V,
positive puts positive Q, on the gate. The positive charge pushes the holes
away from the region under the gate and uncovers some of the negatively-
charged fixed acceptors. Now the electric field points the other way, and
goes from the positive gate charge, terminating on the negative acceptor
charge within the silicon.

SR WAWS
INSNONG!

x

Increasing the voltage

extends the depletion

region further into the
device.

The electric field now extends into the semiconductor. We know from our
experience with the p-n junction that when there is an electric field, there is
a shift in potential, which is represented in the band diagram by bending the
bands. Bending the bands down (as we should moving towards positive
charge) causes the valence band to pull away from the Fermi level near the
surface of the semiconductor. If you remember the expression we had for
the density of holes in terms of E, and E; it is easy to see that indeed, [link],
there is a depletion region (region with almost no holes) near the region
under the gate. (Once Fr - E, gets large with respect to kT, the negative
exponent causes p — 0.)

Equation:

>

s

Le;

ON

>,

SS

SS

SS
uv

2

WU

iA

SS

PFs

Threshold, E; is getting
close to E,.

The electric field extends further into the semiconductor, as more negative
charge is uncovered and the bands bend further down. But now we have to
recall the electron density equation, which tells us how many electrons we
have:

Equation:

Ec—Ef
n= N.e~ kT

A glance at [link] reveals that with this much band bending, E, the
conduction band edge, and E; the Fermi level are starting to get close to one
another (at least compared to kT), which means that n, the electron
concentration, should soon start to become significant. In the situation
represented by [link], we say we are at threshold, and the gate voltage at
this point is called the threshold voltage, Vr.

Now, let's increase V, above V7. Here's the sketch in [link]. Even though we
have increased Vg beyond the threshold voltage, V;, and more positive
charge appears on the gate, the depletion region no longer moves back into
the substrate. Instead electrons start to appear under the gate region, and the
additional electric field lines terminate on these new electrons, instead of on
additional acceptors. We have created an inversion layer of electrons under
the gate, and it is this layer of electrons which we can use to connect the

two n-type regions in our initial device.

J us

Inversion - electrons form
under the gate.

Where did these electrons come from? We do not have any donors in this
material, so they can not come from there. The only place from which
electrons could be found would be through thermal generation. Remember,
in a semiconductor, there are always a few electron hole pairs being
generated by thermal excitation at any given time. Electrons that get created
in the depletion region are caught by the electric field and are swept over to
the edge by the gate. I have tried to suggest this with the electron generation
event shown in the band diagram in the figure. In a real MOS device, we
have the two n-regions, and it is easy for electrons from one or both to "fall"
into the potential well under the gate, and create the inversion layer of
electrons.

Introduction to the MOS Transistor and MOSFETs

Note:This module is adapted from the Connexions modules entitled
Introduction to MOSFETs and MOS Transistor by Bill Wilson.

We now move on to another three terminal device - also called a transistor.
This transistor, however, works on much different principles than does the
bipolar junction transistor of the last chapter. We will now focus on a device
called the field effect transistor, or metal-oxide-semiconductor field effect
transistor or simply MOSFET.

In [link] we have a block of silicon, doped p-type. Into it we have made two
regions which are doped n-type. To each of those n-type regions we attach a
wire, and connect a battery between them. If we try to get some current, J,
to flow through this structure, nothing will happen, because the n-p junction
on the RHS is reverse biased, i.e., the positive lead from the battery going
to the n-side of the p-n junction. If we attempt to remedy this by turning the
battery around, we will now have the LHS junction reverse biased, and
again, no current will flow. If, for whatever reason, we want current to flow,
we will need to come up with some way of forming a layer of n-type
material between one n-region and the other. This will then connect them
together, and we can run current in one terminal and out the other.

p-type silicon

The start of a field effect

transistor.

To see how we will do this, let's do two things. First we will grow a layer of
SiO, (silicon dioxide or silica, but actually refered to as "oxide") on top of
the silicon. To do this the wafer is placed in an oven under an oxygen
atmosphere, and heated to 1100 °C. The result is a nice, high-quality
insulating SiO> layer on top of the silicon). On top of the oxide layer we
then deposit a conductor, which we call the gate. In the "old days" the gate
would have been a layer of aluminum; hence the "metal-oxide-silicon" or
MOS name. Today, it is much more likely that a heavily doped layer of
polycrystalline silicon (polysilicon, or more often just "poly") would be
deposited to form the gate structure. Polysilicon is made from the reduction
of a gas, such as silane (SiH,), [link].

Equation:

The silicon is polycrystalline (composed of lots of small silicon crystallites)
because it is deposited on top of the oxide, which is amorphous, and so it
does not provide a single crystal "matrix" which would allow the silicon to
organize itself into one single crystal. If we had deposited the silicon on top
of a single crystal silicon wafer, we would have formed a single crystal
layer of silicon called an epitaxial layer. This is sometimes done to make
structures for particular applications. For instance, growing a n-type
epitaxial layer on top of a p-type substrate permits the fabrication of a very
abrupt p-n junction.

Note:Epitaxy, is a transliteration of two Greek words epi, meaning "upon",
and taxis. meaning "ordered". Thus an epitaxial layer is one that follows
the order of the substrate on which it is grown.

Now we can go back now to our initial structure, shown in [link], only this
time we will add an oxide layer, a gate structure, and another battery so that
we can invert the region under the gate and connect the two n-regions
together. Well also identify some names for parts of the structure, so we will
know what we are talking about. For reasons which will be clear later, we
call the n-region connected to the negative side of the battery the source,
and the other one the drain. We will ground the source, and also the p-type
substrate. We add two batteries, V;, between the gate and the source, and
Vas between the drain and the source.

41] 1|IF-*
IKV

Vgs

{I

p-type substrate

Biasing a MOSFET transistor

It will be helpful if we also make another sketch, which gives us a
perspective view of the device. For this we strip off the gate and oxide, but
we will imagine that we have applied a voltage greater than Vr to the gate,
so there is a n-type region, called the channel which connects the two. We
will assume that the channel region is Z long and W wide, as shown in
[link].

BU dx

The inversion channel and its
resistance.

Next we want to take a look at a little section of channel, and find its
resistance a(R), when the little section is O(a) long, [link].
Equation:

We have introduced a slightly different form for our resistance formula
here. Normally, we would have a simple o in the denominator, and an area
A, for the cross-sectional area of the channel. It turns out to be very hard to
figure out what that cross sectional area of the channel is however. The
electrons which form the inversion layer crowd into a very thin sheet of
surface charge which really has little or no thickness, or penetration into the
substrate.

If, on the other hand we consider a surface conductivity (units: simply
mhos), o, [link], then we will have an expression which we can evaluate.
Here, jz; is a surface mobility, with units of cm2/V.sec, that is the quantity
which represented the proportionality between the average carrier velocity
and the electric field, [link] and [link].

Equation:
Os = HsQ chan
Equation:
v=pE
Equation:
qT
an

The surface mobility is a quantity which has to be measured for a given
system, and is usually just a number which is given to you. Something
around 300 cm?/V.sec is about right for silicon. Q chan is called the surface
charge density or channel charge density and it has units of Coulombs/cm?.
This is like a sheet of charge, which is different from the bulk charge
density, which has units of Coulombs/cm2. Note that:

Equation:
2 Coul
cm Coulombs __ sec
Volt sec cm? “Volt
ees
—~ VY
= mbhos

It turns out that it is pretty simple to get an expression for Q chan, the surface
charge density in the channel. For any given gate voltage V,,, we know that
the charge density on the gate is given simply as:

Equation:

Qo = Cox Ves

However, until the gate voltage V,., gets larger than Vr we are not creating
any mobile electrons under the gate, we are just building up a depletion
region. We'll define Q as the charge on the gate necessary to get to
threshold. Q-7 = CoxVr. Any charge added to the gate above Q7 is
matched by charge @ chan in the channel. Thus, it is easy to say: [link] or
[link].

Equation:

OQ) saan = Q, _ Qr

Equation:

Q chan = Cox (V, _ Vr)

Thus, putting [link] and [link] into [link], we get:
Equation:

_ al(x)
7 LsCox (Vas = Vr)W

If you look back at [link], you will see that we have defined a current Ig
flowing into the drain. That current flows through the channel, and hence
through our little incremental resistance a(R), creating a voltage drop
a(V.) across it, where V; is the channel voltage, [link].

Equation:

al(V.(«))

Iydl(R)
Izdl(x)
[sCox(Ves—Vr)W

Let's move the denominator to the left, and integrate. We want to do our
integral completely along the channel. The voltage on the channel V,(z)
goes from 0 on the left to Vg, on the right. At the same time, z is going
from 0 to L. Thus our limits of integration will be 0 and Vg, for the voltage
integral Q(V.(x)) and from 0 to L for the z integral a/(z).

Equation:

Vas LT
i [sCox (Ves - Vr Wd Ve = / Igdz
0 0

Both integrals are pretty trivial. Let's swap the equation order, since we
usually want J/g as a function of applied voltages.
Equation:

IgL = [LsCoxW (Vas ad Vr) Vas

We now simply divide both sides by Z, and we end up with an expression
for the drain current Jg, in terms of the drain-source voltage, Vg;, the gate

voltage V,, and some physical attributes of the MOS transistor.
Equation:

sCoxW
i= ( eet Vee v1) Vai

Light Emitting Diode
Light Emitting Diode

Note:This module is adapted from the Connexions module entitled Light
Emitting Diode by Bill Wilson.

Let's talk about the recombining electrons for a minute. When the electron
falls down from the conduction band and fills in a hole in the valence band,
there is an obvious loss of energy. The question is; where does that energy
go? In silicon, the answer is not very interesting. Silicon is what is known
as an indirect band-gap material. What this means is that as an electron
goes from the bottom of the conduction band to the top of the valence band,
it must also undergo a significant change in momentum. This all comes
about from the details of the band structure for the material, which we will
not concern ourselves with here. As we all know, whenever something
changes state, we must still conserve not only energy, but also momentum.
In the case of an electron going from the conduction band to the valence
band in silicon, both of these things can only be conserved if the transition
also creates a quantized set of lattice vibrations, called phonons, or "heat".
Phonons posses both energy and momentum, and their creation upon the
recombination of an electron and hole allows for complete conservation of
both energy and momentum. All of the energy which the electron gives up
in going from the conduction band to the valence band (1.1 eV) ends up in
phonons, which is another way of saying that the electron heats up the
crystal.

In some other semiconductors, something else occurs. In a class of
materials called direct band-gap semiconductors, the transition from
conduction band to valence band involves essentially no change in
momentum. Photons, it turns out, possess a fair amount of energy (several
eV/photon in some cases) but they have very little momentum associated
with them. Thus, for a direct band gap material, the excess energy of the
electron-hole recombination can either be taken away as heat, or more
likely, as a photon of light. This radiative transition then conserves energy

and momentum by giving off light whenever an electron and hole
recombine. This gives rise to the light emitting diode (LED). Emission of a
photon in an LED is shown schematically in [link].

OVEOQO
Eg ade
Ey
OHOOVE
© © @°

Radiative recombination in a
direct band-gap semiconductor.

It was Planck who postulated that the energy of a photon was related to its
frequency by a constant, which was later named after him. If the frequency
of oscillation is given by the Greek letter "nu" (v), then the energy of the
photon is just given by, [link], where h is Planck's constant, which has a
value of 4.14 x 10° eV.sec.

Equation:

E=hv

When we talk about light it is conventional to specify its wavelength, A,
instead of its frequency. Visible light has a wavelength on the order of
nanometers, e.g., red is about 600 nm, green about 500 nm and blue is in
the 450 nm region. A handy "rule of thumb" can be derived from the fact
that c = Av, where c is the speed of light (3 x 10° m/sec or 3 x 10!” nm/sec,
[link].

Equation:

A(nm) = ev)

1242
E(eV)

Thus, a semiconductor with a 2 eV band-gap should give off light at about
620 nm (in the red). A 3 eV band-gap material would emit at 414 nm, in the
violet. The human eye, of course, is not equally responsive to all colors
({link]). The materials which are used for important light emitting diodes
(LEDs) for each of the different spectral regions are also shown in [link].

lnva!

I[EVE
IME

350 400 450 500 550 600 650 700 750
wavelength in nanometers

Relative response of the human
eye to various colors.

It is worth noting that a number of the important LEDs are based on the
GaAsP system. GaAs is a direct band-gap semiconductor with a band gap
of 1.42 eV (in the infrared). GaP is an indirect band-gap material with a
band gap of 2.26 eV (550 nm, or green). Both As and P are group V
elements. (Hence the nomenclature of the materials as III-V (or 13-15)
compound semiconductors.) We can replace some of the As with P in GaAs
and make a mixed compound semiconductor GaAsj_,P,. When the mole
fraction of phosphorous is less than about 0.45 the band gap is direct, and
so we can "engineer" the desired color of LED that we want by simply
growing a crystal with the proper phosphorus concentration! The properties
of the GaAsP system are shown in [link]. It turns out that for this system,
there are actually two different band gaps, as shown in [link]. One is a
direct gap (no change in momentum) and the other is indirect. In GaAs, the
direct gap has lower energy than the indirect one (like in the inset) and so
the transition is a radiative one. As we start adding phosphorous to the
system, both the direct and indirect band gaps increase in energy. However,
the direct gap energy increases faster with phosphorous fraction than does

the indirect one. At a mole fraction x of about 0.45, the gap energies cross
over and the material goes from being a direct gap semiconductor to an
indirect gap semiconductor. At x = 0.35 the band gap is about 1.97 eV (630
nm), and so we would only expect to get light up to the red using the
GaAsP system for making LED's. Fortunately, people discovered that you
could add an impurity (nitrogen) to the GaAsP system, which introduced a
new level in the system. An electron could go from the indirect conduction
band (for a mixture with a mole fraction greater than 0.45) to the nitrogen
site, changing its momentum, but not its energy. It could then make a direct
transition to the valence band, and light with colors all the way to the green
became possible. The use of a nitrogen recombination center is depicted in
the [link].

Conduction Band

=
z a
: 4
]
= —
5 =~
= =
3
o o2 O4 06 0.8 1
GaAs GaP

Mole Fraction Phosphorous

Band gap for the GaAsP system

Energy

hv
\

Addition of a
nitrogen
recombination
center to indirect
GaAspP.

4
kaa

Momentum

If we want colors with wavelengths shorter than the green, we must
abandon the GaAsP system and look for more suitable materials. A
compound semiconductor made from the II-VI elements Zn and Se make up
one promising system, and several research groups have successfully made
blue and blue-green LEDs from ZnSe. SiC is another (weak) blue emitter
which is commercially available on the market. Recently, workers at a tiny,
unknown chemical company stunned the "display world" by announcing
that they had successfully fabricated a blue LED using the II-V material
GaN. A good blue LED was the "holy grail" of the display and CD ROM
research community for a number of years. Obviously, adding blue to the
already working green and red LED's completes the set of 3 primary colors
necessary for a full-color flat panel display. Furthermore, using a blue LED
or laser in a CD ROM would more than quadruple its data capacity, as bit
diameter scales as A, and hence the area as A2.

Polymer Light Emitting Diodes

This module was developed as part of a Rice University course CHEM496:
Chemistry of Electronic Materials. This module was prepared with the assistance
of Pui Yee Hung.

Introduction

In 1990, electroluminescent (EL) from conjugated polymers was first reported by
Burroughes et al. of Cambridge University. A layer of poly(para-
phenylenevinylene) (PPV) was sandwiched between layers of indium tin oxide
(ITO) and aluminum. When this device is under a 14 V dc bias, the PPV emits a
yellowish-green light with a quantum efficiency of 0.05%. This report attracted a
lot of attention, because the potential that polymer light emitting diodes (LEDs)
could be inexpensively mass produced into large area display area. The processing
steps in making polymer LEDs are readily scaleable. The industrial coating
techniques is well developed to mass produce polymer layers of 100 nm
thickness, and the device could be patterned onto large surface area by pixellation
of metal.

Since the initial discovery, and increasing amount of researches has been
performed, and significant progress has been made. In 1990 the polymer LED
only emitted yellowish green color, now the emission color ranged from deep blue
to near infra red. The efficiency of the multi-layer polymer LED even reached a
quantum efficiency of >4% and the operating voltage has been reduced
significantly. In term of efficiency, color selection and operating voltage, polymer
LEDs have attained adequate levels for commercialization. But there are
reliability problems that are symptomatic of any organic devices.

Device physics and materials science of polymer LEDs

A schematic diagram of a polymer LED is shown in [link]. A polymer LED can
be divided into three different components:

A. Anode: the hole supplier, made of metal of high working function. Examples
of the common anode are indium tin oxide (ITO), gold etc. The anode is
usually transparent so that light can be emitted through.

B. Cathode: the electron supplier, made of metal of low working function.
Examples of the common cathode are aluminum or calcium.

C. Polymer: made of conjugated polymer film with thickness of 100 nm.

Cathode (aluminum)

Anode (ITO)

Substrate
(glass)

Emitted light

Schematic set-up of polymer LED.

When a polymer LED is under a direct current (dc) bias, holes are injected from
the anode (ITO) and electrons are injected from the cathode (aluminum). Under
the influences of the electrical field, the electrons and holes will migrate toward
each other. When they recombine in the conjugated polymer layer, a bound
excited states (excitons) will be formed. Some of the excitons (singlets) then
decays in the conjugated polymer layer to emit light through the transparent
substrates (glass). The emission color will be depended on the energy gap of the
polymers. There is energy gap in a conjugated polymer because the m electron are
not completely delocalized over the entire polymer chain. Instead there are
alternate region in the polymer chain that has a higher electron density ({link]a).
The chain length of this region is about 15-20 multiple bonds. The emission color
can be controlled by tuning this energy band gap ((link]|b). It shows that bond
alternation limits the extent of delocalization. [link] summarizes the structure and
emission color of some common conjugated polymers.

(a)

Alternation of bond lengths along a conjugated polymer
chain (a) results in a material with properties of a large
band gap semiconductor (b), where CB is the conductive
band gap, and VB is the valence band, and E, is the band

Polymer

PA

PDA

PPP

gap.

Chemical name

trans-
polyacetylene

polydiacetylene

poly(para-
phenylene)

Structure

m™1-Tt*
energy
gap
(eV)

1.5

Emission
peak
(nm)

600

465

PPV Poly(para- Oat 2.5 565
phenylenevinylene) : (green)
poly(2,5-dialkoxy- - 09

RO-PPV __p- +, (blue) 980
phenlyenevinylen) - :

; s 2.0

PT polythiophene an Th (red)
Poly(3 ss “ye | 20

O yi 7} n .

ea alkythiophene) : (red) on
Poly(2,5- :

raN thiophenevinylene) TUT, ae

i!
PPy Polypyrrole an@r at 3.1
PAni Polyaniline so eo 3.2

Example of common conjugated polymers.

Approaches to improve the efficiency
Efficiency for any LED is defined:

Next = Desc ~ int

where Nexis the external quantum efficiency, nj; is the internal efficiency
(represents the fraction of injected carrier, usually electron, that is converted to
photon), and n,,, is the escape efficiency (represent fraction of photons that can
reach to the outside).

The most common way to improve the internal efficiency is to balance the
number of electrons and holes which arrives at the polymer layer. Originally, there
are more holes than electron that arrive of the polymer layer because conjugated
polymers have a higher electron affinity, and as a consequence will favor the
transport of hole than electron. There are two ways to maintains the balance:

1. Match the work function of electrode with the electron affinity and ionization
potential of the polymer.

2. Tune the polymer’s electron affinity and ionization potential to match the
work function of the electrode.

The escape efficiency is also important because a polymer LED is made up of
layers of materials that have different refractive index, and some of the photon
generated from the excition may be reflected at the boundary and trapped inside
the device.

Improvement in internal quantum efficiency using low working function
cathode

Conjugated polymer is electron rich, the mobility for hole is higher than electron,
and more holes will arrive in the polymer layer than electrons. One way to
increase the population of the electron is to use a lower working function metal as
cathode. Braun and Heeger have replaced the aluminum cathode with calcium
results in improved internal efficiency by a factor of ten, to 0.1%. This approach
is direct and fast but low working function electrode like calcium will be oxidized
easily and shorten the devices’ life.

Improvement in internal quantum efficiency using multiple polymer layers

A layer of poly[2,5-di(hexyloxy)cyanoterephthalylidene] (CN-PPYV, [link]) is
coated on top of PPV to improve the transport and recombination of electron and
holes ({link]).

CsH)30

Structure of CN-PPV.

Cathode (aluminum)

NOUN UN UN UN UN UN UN UN UN UN UN UNOS UN UN UN UN UN
C4 OL OOOO LOE LOSS

Substrate
(glass)

Anode (ITO)

Emitted light

Schematic representation of a CN-PPV and PPV
multi-layer LED.

The nitrile group in the CN-PPV has two effect on the polymer.

1. It increases the electron affinity so electrons can travel more efficient from
the aluminum to the polymer layer. And metal of relative high working
function like aluminum and gold can be now be used as cathode instead of
calcium.

2. It increases the binding energy of the occupied m and unoccupied m* state but
maintain a similar m-1* gap. So when the PPV and CN-PPV is placed

together, holes and electron will be confined at the heterojunction.

The resulting energy levels are shown in [link].

PPV CN-PPV

Schematic energy-level diagram for a PPV and
CN-PPV under foreword bias. Adapted from N. C.
Greenham, S. C. Maratti, D. D. C. Bradley, R. H.
Friend, and A. B. Holmes, Nature,1993, 365, 62.

The absolute energies of levels are not known accurately, but the diagram show
the relative position of the HOMO and LUMO levels in the polymers, and the
Fermi levels of the various possible metal contacts, the differences in electron
affinity (AEA) and ionization potential (AIP) between PPV and CN-PPV are also
shown ([link]).

At the polymers interface there is a sizable offset in the energies of HOMO and
LUMO of PPV and CN-PPY, the holes transported from the ITO and the electrons
transport from the aluminum will be confined in the heterojunction. The local
charge density will be sufficiently high to ensure the holes and electrons will pass
within a collision capture radius. This set-up increases the chance for an electrons
to combine with holes to form an excition. In addition, the emission will be close
to the junction, far away from the electrode junction which will quench the singlet
excitions. The result is that a multi-layers LED has an internal quantum efficiency

of 10% and external quantum efficiency (for light emitted in foreword direction)
of 25%.

Based on this approach, a couple of polymers have been developed or modified to
produce the desirable emission color and processing property. The drawback of
this method is that desirable properties may not be commentary to each other. For
example, in MEH-PPV an alkoxy side group (RO) is introduced to PPV so that it
can be dissolved in organic solvent. But the undesirable effect is that MEH-PPV is
less thermally stable. Moreover in multiple layers LEDs, different polymer layers
have different refractive indices and a fraction of the photons will undergo total
internal reflection at the refractive boundaries and cannot escape as light. This
problem can be overcome by Febry-Pert microcavity structure.

Improvement in external quantum efficiency using microcavity

Fabry-Perot resonant structures are also used in inorganic LED, and are is based
on Fermi’s golden rule:

K,~ | <M > | tw

where M (the matrix element of the perturbation between final and initial states)
depends on the nature of the material, and r;,) can be altered by changing the
density of various density states, e.g. using a luminescent thin films to select
certain value of V.

In building a microcavity for a polymer LED, the polymer is placed between two
mirrors. ((link]), in which one of the mirrors is made up of aluminum, the other
mirror (a Bragg Mirror) is form by epitaxial multilayer stacks of Si,Ny and SiOp.

Aluminum mirror

LEKI RAL AL
REEL Polymer
AN

Si,N,
SiO,

Schematic set-up of micro-cavity.

Improvement in internal quantum efficiency: doping of polymer

Doping is a process that creates carrier by purposely introducing impurities and is
very popular method in the semiconductor industry. However, this technique was
not used in polymer LED until 1995, when a co-polymer polystyrene-poly(3-
hexylthiphene) (PS-P3HT) was doped with FeCl; Doping of MEH-PPV with
iodine has improved the efficiency by 200% and the polymer LED can be
operated under both forward and reverse bias ([link]). The doping is accomplsihed
by mixing 1 wt% MEH-PPV with 0.2 wt% Ip. The molar ratio of MEH-PPV to Ip
is 5:1. That is a huge “doping “ ratio when you compare the doping concentration
in the semiconductor.

Un-doped Doped
Turn on voltage (V) 10 foreword 5, reversed 12

External efficiency (%) 4x 104 8x 10°

Results of iodine doping of an Al/MEH-PPV/ITO-based LED.

Polymer LEDs on a silicon substrate: an application advantage over
inorganic LEDs

In the initial research polymer LEDs were in direct competition with the inorganic
LEDs and tried to achieve the existing LED standard. This is a difficult task as
polymer LEDs have a lower long term stability. However, there are some
applications in which polymer LEDs have a clear advantage over their more
traditional inorganic analogs. One of these is to incorporate LEDs with the silicon
integrated circuits for inter-chip communication.

It is difficult to build inorganic LEDs on a silicon substrate, because of the
thermal stress developing between the inorganic LED (usually a HI-V based
device) and the silicon interface. But polymer LEDs offer a solution, since
polymers can be easily spin-coated on the silicon. The operating voltage of
polymer LED is less than 4 V, and the turn on voltage can be as low as 2 V.
Together with a switching time of less than 50 ns, make polymer LED a perfect
candidate.

Reliability and degradation of polymer LEDs

In terms of the efficiency, color selection, and driving voltage, polymer LED have
attained adequate level for commercialization. However, the device lifetime is still
far from satisfactory. Research into understanding the reliability and degredation
mechanisms of polymer LEDs has generally been divided into two area:

1. Photo-degradation of polymer.
2. Interface degradation.

Polymer photo degradation

Photoluminescece (PL) studies of the photo-oxidation of PPV have been
undertaken, since it is believes that EL is closely related with PL.

It was found that there is a rapid decay in emission when PPV is exposed to
oxygen. Using time resolved FTIR spectroscopy an increase in the carbonyl signal
and a decrease in C=C signal with time ((link]). It was suggested that the carbonyl

group has a strong electron affinity level to charge transfer between molecules
segment in the polymer, thereby dissociating the excition and quenching the PL.

Change in
absorbance
(arb units)

1800 1700 1600 1500 , 1000 900

Frequency (cm!)

FTIR as a function of photo-oxidation of
PPV. Adapted from M. Yan, L. J. Rothberg,
F, Papadimitrakopoulos, M. E. Galvin and T.

M. Miller, Phys. Rev. Lett., 1994, 73, 744.

Similar research was performed by Cumpston and Jensen using BCHA-PPV and
P30T ([link]) and exposing them to dry air in UV irradiation. In BCHA-PPV,
there is an increase in carbonyl signal with time, while the P3OT remain intact. A
mechanism proposed for the degradation of BCHA-PPV involves the transfer of
energy from the excited triplet state of the PPV to oxygen to from singlet oxygen
which attack the vinyl double bond in the PPV backbone. And P3O0T dose not has
vinyl bond so it can resist the oxidation .

(a) (b)

Structure of (a) BCHA-PPV and (b) P30T.

The research described above was all performed on polymer thin films deposited
on an inert surface. The presence of cathode and anode may also affect the
oxidation mechanism. Scott et al. have taken IR spectra from a MEH-PPV LED in
the absence of oxygen. They obtained similar result as in Yan et al., however, a
decrease in ITO’s oxygen signal was noticed suggesting that the ITO anode acts
like a oxygen reservoir and supplies the oxygen for the degradation process.

Polymer LED interface degradation

There are few interface degredation studies in polymer LEDs. One of them by
Scott et al. took SEM image of the cathode from a failed polymer LED. The
polymer LED used ITO as the anode, MEH-PPV as the polymer layer, and an
aluminum calcium alloy as cathode. SEM images showed “craters” formed in the
cathode. The craters are formed when the cathode metal is melted and pull away
from the polymer layer. It was suggested that a high current density will generate
heat and result in local hot spot. The temperature in the hot spot is high enough to
melt the cathode. And when it melt, it will pull away from the polymer. This
process will decrease the effective cathode area, and reduce the luminescence
gradually.

Bibliography

D. R. Baigent, N. C. Greenham, J. Gruner, R. N. Marks, R. H. Friends, S. C.
Moratti, and A. B. Holmes, Synth.Met., 1994, 67, 3.

B. H. Cumpston and K. F. Jensen, Synth. Met., 1995, 73, 195.

J. H. Burroughes, D. D. C. Bradley, A. R. Brown, R. N. Marks, K. Mackay,
R. H. Friend, P. L. Burns, and A. B. Holmes, Nature, 1990, 347, 539.

N. C. Greenham, S. C.Maratti, D. D. C. Bradley, R. H. Friend, and A. B.
Holmes, Nature, 1993, 365, 628.

J. Gruner, F. Cacialli, I. D. W. Samuel, R. H. Friend, Synth. Met, 1996, 76,
197.

M. Herold, J. Gmeiner, W. Riess, and M. Schwoerer, Synth. Met., 1996, 76,
109.

R. H. Jordan, A. Dodabalapur, L. J. Rothberg, and R. E. Slusher, Proceeding
of SPIE, 1997, 3002, 92.

I. D. Parker and H. H. Kim, Appl. Phys. Lett., 1994, 64, 1774.

J. C. Scott, J. Kaufman, P. J. Brock, R. DiPietro, J. Salem, and J. A. Goitia, J.
Appl. Phys., 1996, 79, 2745.

M. S. Weaver, D. G. Lidzaey, T. A. Fisher, M. A. Pate, D. O’Brien, A.
Bleyer, A. Tajbakhsh, D. D. C. Bradley, M. S. Skolnick, and G. Hill, Thin
solid Films, 1996, 273, 39.

M. Yan, L. J.Rothberg, F. Papadimitrakopoulos, M. E. Galvin, and T. M.
Miller, Phys. Rev. Lett., 1994, 73, 744.

Laser
LASER

Note:This module is adapted from the Connexions module entitled LASER
by Bill Wilson.

What is the difference between an LED and a solid state laser? There are
some differences, but both devices operate on the same principle of having
excess electrons in the conduction band of a semiconductor, and arranging
it so that the electrons recombine with holes in a radiative fashion, giving
off light in the process. What is different about a laser? In an LED, the
electrons recombine in a random and unorganized manner. They give off
light by what is known as spontaneous emission, which simply means that
the exact time and place where a photon comes out of the device is up to
each individual electron, and things happen in a random way.

There is another way in which an excited electron can emit a photon
however. If a field of light (or a set of photons) happens to be passing by an
electron in a high energy state, that light field can induce the electron to
emit an additional photon through a process called stimulated emission. The
photon field stimulates the electron to emit its energy as an additional
photon, which comes out in phase with the stimulating field. This is the big
difference between incoherent light (what comes from an LED or a
flashlight) and coherent light which comes from a laser. With coherent
light, all of the electric fields associated with each phonon are all exactly in
phase. This coherence is what enables us to keep a laser beam in tight
focus, and to allow it to travel a large distance without much divergence or
spreading out.

So how do we restructure an LED so that the light is generated by
stimulated emission rather than spontaneous emission? Firstly, we build
what is called a heterostructure. All this means is that we build up a
sandwich of somewhat different materials, with different characteristics. In
this case, we put two wide band-gap regions around a region with a

narrower band gap. The most important system where this is done is the
AlGaAs/GaAs system. A band diagram for such a set up is shown in [link].
AlGaAs (pronounced "Al-Gas") has a larger band-gap then does GaAs. The
potential "well" formed by the GaAs means that the electrons and holes will
be confined there, and all of the recombination will occur in a very narrow
strip. This greatly increases the chances that the carriers can interact, but we
still need some way for the photons to behave in the proper manner. [Link] is
a diagram of what a typical diode might look like. We have the active GaAs
layer sandwich in-between the two heterostructure confinement layers, with
a contact on top and bottom. On either end of the device, the crystal has
been "cleaved" or broken along a crystal lattice plane. This results in a
shiny "mirror-like" surface, which will reflect photons. The back surface
(which we can not see here) is also cleaved to make a mirror surface. The
other surfaces are purposely roughened so that they do not reflect light.
Now let us look at the device from the side, and draw just the band diagram
for the GaAs region ({link]). We start things off with an electron and hole
recombining spontaneously. This emits a photon which heads towards one
of the mirrors. As the photon goes by other electrons, however, it may cause
one of them to decay by stimulated emission. The two (in phase) photons
hit the mirror and are reflected and start back the other way . As they pass
additional electrons, they stimulate them into a transition as well, and the
optical field within the laser starts to build up. After a bit, the photons get
down to the other end of the cavity. The cleaved facet, while it acts like a
mirror, is not a perfect one. Some light is not reflected, but rather "leaks";
though, and so becomes the output beam from the laser. The details of
finding what the ratio of reflected to transmitted light is will have to wait
until later in the course when we talk about dielectric interfaces. The rest of
the photons are reflected back into the cavity and continue to stimulate
emission from the electrons which continue to enter the gain region because
of the forward bias on the diode.

n-AlGaAs GaAs p-AlGaAs

The band diagram for a double
heterostructure GaAs/AlGaAs
laser.

A schematic diagram of a
typical laser diode.

Build up of a photon
field in a laser diode.

In reality, the photons do not move back and forth in a big "clump" as we
have described here, rather they are distributed uniformly along the gain
region ([link]). The field within the cavity will build up to the point where
the loss of energy by light leaking out of the mirrors just equals the rate at
which energy is replaced by the recombining electrons.

OOQOOQ90 O00 OO

y yy

g Qin

y ve
OOOO OOOO 0000

Output coupling in a diode
laser.

Solar Cells

Note:This module is adapted from the Connexions module entitled Solar
Cells by Bill Wilson.

Now let us look at the opposite process of light generation for a moment.
Consider the following situation where we have just a plain old normal p-n
junction, only now, instead of applying an external voltage, we imagine that
the junction is being illuminated with light whose photon energy is greater
than the band-gap ({link]a). In this situation, instead of recombination, we
will get photo-generation of electron hole pairs. The photons simply excite
electrons from the full states in the valence band, and "kick" them up into
the conduction band, leaving a hole behind. This is similiar to the thermal
excitation process. As can be seen from [link]b, this creates excess electrons
in the conduction band in the p-side of the diode, and excess holes in the
valence band of the n-side. These carriers can diffuse over to the junction,
where they will be swept across by the built-in electric field in the depletion
region. If we were to connect the two sides of the diode together with a
wire, a current would flow through that wire as a result of the electrons and
holes which move across the junction.

A schematic representation of a p-n diode under
illlumination.

Which way would the current flow? A quick look at [link]c shows that
holes (positive charge carriers) generated on the n-side will float up to the
p-side as they go across the junction. Hence positive current must be
coming out of the anode, or p-side of the junction. Likewise, electrons
generated on the p-side will fall down the junction potential, and come out
the n-side, but since they have negative charge, this flow represents current
going into the cathode. We have constructed a photovoltaic diode, or solar
cell. [link] is a picture of what this would look like schematically. We might
like to consider the possibility of using this device as a source of energy, but
the way we have things set up now, since the voltage across the diode is
zero, and since power equals current times voltage, we see that we are
getting nada from the cell. What we need, obviously, is a load resistor, so
let's put one in. It should be clear from [link] that the photo current flowing
through the load resistor will develop a voltage which it biases the diode in
the forward direction, which, of course will cause current to flow back into

the anode. This complicates things, it seems we have current coming out of
the diode and current going into the diode all at the same time! How are we
going to figure out what. what is going on?

=

photon flux

IK

Schematic representation
of a photovoltaic cell.

+

€ Vv
= out

Photovoltaic cell
with a load
resistor.

The answer is to make a model. The current which arises due to the photon
flux can be conveniently represented as a current source. We can leave the
diode as a diode, and we have the circuit shown in [link]. Even though we
show I,,; coming out of the device, we know by the usual polarity
convention that when we define V,,; as being positive at the top, then we
should show the current for the photovoltaic, I,y as current going into the
top, which is what was done in [Link]. Note that Ipy = Idiode - Iphoto, $0 all we
need to do is to subtract the two currents; we do this graphically in [link].

Note that we have numbered the four quadrants in the I-V plot of the total
PV current. In quadrant I and III, the product of I and V is a positive
number, meaning that power is being dissipated in the cell. For quadrant IT
and IV, the product of I and V is negative, and so we are getting power from
the device. Clearly we want to operate in quadrant IV. In fact, without the
addition of an external battery or current source, the circuit, will only run in
the IV'th quadrant. Consider adjusting R;,, the load resistor from 0 (a short)
to co (an open). With R;,, we would be at point A on [link]. As Rj, starts to
increase from zero, the voltage across both the diode and the resistor will
start to increase also, and we will move to point B, say. As Ry, gets bigger
and bigger, we keep moving along the curve until, at point C, where Ry, is
an open and we have the maximum voltage across the device, but, of

course, no current coming out!
lout ————_»

A model of a PV cell.

| diode | photo

| pv

Combining the diode and
the current source.

Power is VJ so at B for instance, the power coming out would be
represented by the area enclosed by the two dotted lines and the coordinate
axes. Someplace about where I have point B would be where we would be
getting the most power out of out solar cell.

[link] shows you what a real solar cell would look like. They are usually
made from a complete wafer of silicon, to maximize the usable area. A
shallow (0.25 tm) junction is made on the top, and top contacts are applied
as stripes of metal conductor as shown. An anti-reflection (AR) coating is
applied on top of that, which accounts for the bluish color which a typical
solar cell has ({link]).

Solar Cell Wafer

top contact AR coating

a

back contact
Side View

A schematic diagram of a real
solar cell.

A solar cell showing the blue
tint due to the AR coating.

The solar power flux on the earth's surface is (conveniently) about 1 kW/m?
or 100 mW/cm-. So if we made a solar cell from a 4 inch diameter wafer
(typical) it would have an area of about 81cm? and so would be receiving a
flux of about 8.1 Watts. Typical cell efficiencies run from about 10% to
maybe 15% unless special (and costly) tricks are made. This means that we
will get about 1.2 Watts out from a single wafer. Looking at B on 2.59 we
could guess that Vout will be about 0.5 to 0.6 volts, thus we could expect to
get maybe around 2.5 amps from a 4 inch wafer at 0.5 volts with 15%
efficiency under the illumination of one sun.

Properties of Gallium Arsenide

Gallium: the element

The element gallium was predicted, as eka-aluminum, by Mendeleev in
1870, and subsequently discovered by Lecog de Boisbaudran in 1875; in
fact de Boisbaudran had been searching for the missing element for some
years, based on his own independent theory. The first experimental
indication of gallium came with the observation of two new violet lines in
the spark spectrum of a sample deposited on zinc. Within a month of these
initial results de Boisbaudran had isolated 1 g of the metal starting from
several hundred kilograms of crude zinc blende ore. The new element was
named in honor of France (Latin Gallia), and the striking similarity of its
physical and chemical properties to those predicted by Mendeleev ([link])
did much to establish the general acceptance of the periodic Law; indeed,
when de Boisbaudran first stated that the density of Ga was 4.7 g/cm? rather
than the predicted 5.9 g/cm?, Mendeleev wrote to him suggesting that he
redetermine the value (the correct value is 5.904 g/cm?).

Mendeleev's Observed properties of

Property prediction (1871) for gallium (discovered
eka-aluminum, M 1875)

eee ca. 68 69.72

weight

Density,

ee 5.9 5.904
g.cm
Mens Low 29.78

point

Vapor Non-volatile 10° mmHg, 1000 °C
pressure

Valence 3 3
Oxide M,O3 GayO3
Density
of oxide 5.5 5.88
(g/cm?)
: _ shoul d dissolve Ga metal dissolves
Properties slowly in acids and

slowly in acids and

of metal alkalis and be stable in Albalie end Ge Seble ack

alr

Properties M(OH)3 should
of dissolve in both acids
hydroxide and alkalis

Ga(OH)3 dissolves in
both acids and alkalis

M salts will tend to Ga salts readily

form basic salts; the hydrolyze and form basic

sulfate should form salts; alums are known;
Properties alums; M>S3 should be GaS3 can be precipitated
of salts precipitated by H2S or under special conditions

(NH,)2S; anhydrous by H2S or (NH,)2S,

MCl3 should be more anhydrous GaCl3 is more

volatile than ZnCl» volatile than ZnCl.

Comparison of predicted and observed properties of gallium.

Gallium has a beautiful silvery blue appearance; it wets glass, porcelain,
and most other surfaces (except quartz, graphite, and Teflon®) and forms a
brilliant mirror when painted on to glass. The atomic radius and first
ionization potential of gallium are almost identical with those of aluminum
and the two elements frequently resemble each other in chemical properties.
Both are amphoteric, but gallium is less electropositive as indicated by its

lower electrode potential. Differences in the chemistry of the two elements
can be related to the presence of a filled set of 3d orbitals in gallium.

Gallium is very much less abundant than aluminum and tends to occur at
low concentrations in sulfide minerals rather than as oxides, although
gallium is also found associated with aluminum in bauxite. The main source
of gallium is as a by-product of aluminum refining. At 19 ppm of the earth's
crust, gallium is about as abundant as nitrogen, lithium and lead; it is twice
as abundant as boron (9 ppm), but is more difficult to extract due to the lack
of any major gallium-containing ore. Gallium always occurs in association
either with zinc or germanium, its neighbors in the periodic table, or with
aluminum in the same group. Thus, the highest concentrations (0.1 - 1%)
are in the rare mineral germanite (a complex sulfide of Zn, Cu, Ge, and As);
concentrations in sphalerite (ZnS), bauxite, or coal, are a hundred-fold less.

Gallium pnictides

Gallium's main use is in semiconductor technology. For example, GaAs and
related compounds can convert electricity directly into coherent light (laser
diodes) and is employed in electroluminescent light-emitting diodes
(LED's); it is also used for doping other semiconductors and in solid-state
devices such as heterojunction bipolar transistors (HBTs) and high power
high speed metal semiconductor field effect transistors (MESFETs). The
compound MgGa>O, is used in ultraviolet-activated powders as a brilliant
green phosphor used in Xerox copying machines. Minor uses are as high-
temperature liquid seals, manometric fluids and heat-transfer media, and for
low-temperature solders.

Undoubtedly the binary compounds of gallium with the most industrial
interest are those of the Group 15 (V) elements, GaE (E = N, P, As, Sb).
The compounds which gallium forms with nitrogen, phosphorus, arsenic,
and antimony are isoelectronic with the Group 14 elements. There has been
considerable interest, particularly in the physical properties of these
compounds, since 1952 when Welker first showed that they had
semiconducting properties analogous to those of silicon and germanium.

Gallium phosphide, arsenide, and antimonide can all be prepared by direct
reaction of the elements; this is normally done in sealed silica tubes or in a
graphite crucible under hydrogen. Phase diagram data is hard to obtain in
the gallium-phosphorus system because of loss of phosphorus from the bulk
material at elevated temperatures. Thus, GaP has a vapor pressure of more
than 13.5 atm at its melting point; as compared to 0.89 atm for GaAs. The
physical properties of these three compounds are compared with those of
the nitride in [link]. All three adopt the zinc blende crystal structure and are
more highly conducting than gallium nitride.

Property GaN GaP GaAs GaSb
Mens > 1250 (dec) 1350 1240 712
point (°C)
Density

3 ca. 6.1 4.138 9.3176 9.6137
(g/cm”)
Crystal Wiirtzite zinc zinc zinc
structure blende blende blende
Cell dimen. a= 3.187,c= a= a= a=
(A)? 5.186 5.4505 9.6532 6.0959
Renecuve 2.35 3.178 3.666 4.388
index
k (ohm7!cm 9 4an7 10°? - 10
i 10° - 10 102 ia 6-13
Band eae 3.44 2.24 1.424 0.71

(ev)

Physical properties of 13-15 compound semiconductors. a Values given for
300 K. b Dependent on photon energy; values given for 1.5 eV incident
photons. c Dependent on temperature; values given for 300 K.

Gallium arsenide versus silicon

Gallium arsenide is a compound semiconductor with a combination of
physical properties that has made it an attractive candidate for many
electronic applications. From a comparison of various physical and
electronic properties of GaAs with those of Si ({link]) the advantages of
GaAs over Si can be readily ascertained. Unfortunately, the many desirable
properties of gallium arsenide are offset to a great extent by a number of
undesirable properties, which have limited the applications of GaAs based
devices to date.

Properties GaAs Si
Formula weight 144.63 28.09
Crystal structure zinc blende diamond
Lattice constant 5.6532 5.43095
Melting point (°C) 1238 1415
Density (g/cm?) 5.32 2.328
Thermal conductivity (W/cm.K) 0.46 1.5
Band gap (eV) at 300 K 1.424 1.12

Intrinsic carrier conc. (cm™) 1.79 x 10° 1.45 x 10/0

Intrinsic resistivity (ohm.cm) 108 2.3.x 10°

Breakdown field (V/cm) 4x 10° 3x 10°
Minority carrier lifetime (s) 10" 2.5x 10°
Mobility (cm?/V.s) 8500 1500

Comparison of physical and semiconductor properties of GaAs and Si.

Band gap

The band gap of GaAs is 1.42 eV; resulting in photon emission in the infra-
red range. Alloying GaAs with Al to give Al,Ga,_,As can extend the band
gap into the visible red range. Unlike Si, the band gap of GaAs is direct,
i.e., the transition between the valence band maximum and conduction band
minimum involves no momentum change and hence does not require a
collaborative particle interaction to occur. Photon generation by inter-band
radiative recombination is therefore possible in GaAs. Whereas in Si, with
an indirect band-gap, this process is too inefficient to be of use. The ability
to convert electrical energy into light forms the basis of the use of GaAs,
and its alloys, in optoelectronics; for example in light emitting diodes
(LEDs), solid state lasers (light amplification by the stimulated emission of
radiation).

A significant drawback of small band gap semiconductors, such as Si, is
that electrons may be thermally promoted from the valence band to the
conduction band. Thus, with increasing temperature the thermal generation
of carriers eventually becomes dominant over the intentionally doped level
of carriers. The wider band gap of GaAs gives it the ability to remain
‘intentionally’ semiconducting at higher temperatures; GaAs devices are
generally more stable to high temperatures than a similar Si devices.

Carrier density

The low intrinsic carrier density of GaAs in a pure (undoped) form
indicates that GaAs is intrinsically a very poor conductor and is commonly
referred to as being semi-insulating. This property is usually altered by
adding dopants of either the p- (positive) or n- (negative) type. This semi-
insulating property allows many active devices to be grown on a single
substrate, where the semi-insulating GaAs provides the electrical isolation
of each device; an important feature in the miniaturization of electronic
circuitry, i.e., VLSI (very-large-scale-integration) involving over 100,000
components per chip (one chip is typically between 1 and 10 mm square).

Electron mobility

The higher electron mobility in GaAs than in Si potentially means that in
devices where electron transit time is the critical performance parameter,
GaAs devices will operate with higher response times than equivalent Si
devices. However, the fact that hole mobility is similar for both GaAs and
Si means that devices relying on cooperative electron and hole movement,
or hole movement alone, show no improvement in response time when
GaAs based.

Crystal growth

The bulk crystal growth of GaAs presents a problem of stoichiometric
control due the loss, by evaporation, of arsenic both in the melt and the
growing crystal (> ca. 600 °C). Melt growth techniques are, therefore,
designed to enable an overpressure of arsenic above the melt to be
maintained, thus preventing evaporative losses. The loss of arsenic also
negates diffusion techniques commonly used for wafer doping in Si
technology; since the diffusion temperatures required exceed that of arsenic
loss.

Crystal Stress

The thermal gradient and, hence, stress generated in melt grown crystals
have limited the maximum diameter of GaAs wafers (currently 6" diameter
compared to over 12" for Si), because with increased wafer diameters the
thermal stress generated dislocation (crystal imperfections) densities
eventually becomes unacceptable for device applications.

Physical strength

Gallium arsenide single crystals are very brittle, requiring that considerably
thicker substrates than those employed for Si devices.

Native oxide

Gallium arsenide's native oxide is found to be a mixture of non-
stoichiometric gallium and arsenic oxides and elemental arsenic. Thus, the
electronic band structure is found to be severely disrupted causing a
breakdown in 'normal' semiconductor behavior on the GaAs surface. As a
consequence, the GaAs MISFET (metal-insulator-semiconductor-field-
effect-transistor) equivalent to the technologically important Si based
MOSFET (metal-oxide-semiconductor-field-effect-transistor) is, therefore,
presently unavailable.

The passivation of the surface of GaAs is therefore a key issue when
endeavoring to utilize the FET technology using GaAs. Passivation in this
discussion means the reduction in mid-gap band states which destroy the
semiconducting properties of the material. Additionally, this also means the
production of a chemically inert coating which prevents the formation of
additional reactive states, which can effect the properties of the device.

Bibliography

e S.K. Ghandhi, VLSI Fabrication Principles: Silicon and Gallium
Arsenide. Wiley-Interscience, New York, (1994).

e Properties of Gallium Arsenide. Ed. M. R. Brozel and G. E. Stillman.
3rd Ed. Institution of Electrical Engineers, London (1996).

Synthesis and Purification of Bulk Semiconductors

Introduction

The synthesis and purification of bulk polycrystalline semiconductor
material represents the first step towards the commercial fabrication of an
electronic device. This polycrystalline material is then used as the raw
material for the formation of single crystal material that is processed to
semiconductor wafers. The strong influence on the electric characteristics of
a semiconductors exhibited by small amounts of some impurities requires
that the bulk raw material be of very high purity (> 99.9999%). Although
some level of purification is possible during the crystallization process it is
important to use as high a purity starting material as possible. While a wide
range of substrate materials are available from commercial vendors, silicon
and GaAs represent the only large-scale commercial semiconductor
substrates, and thus the discussion will be limited to the synthesis and
purification of these materials.

Silicon

Following oxygen (46%), silicon (L. silicis flint) is the most abundant
element in the earth's crust (28%). However, silicon does not occur in its
elemental form, but as its oxide (SiO>) or as silicates. Sand, quartz,
amethyst, agate, flint, and opal are some of the forms in which the oxide
appears. Granite, hornblende, asbestos, feldspar, clay and mica, etc. are a
few of the numerous silicate minerals. With such boundless supplies of the
raw material, the costs associated with the production of bulk silicon is not
one of abstraction and conversion of the oxide(s), but of purification of the
crude elemental silicon. While 98% elemental silicon, known as
metallurgical-grade silicon (MGS), is readily produced on a large scale, the
requirements of extreme purity for electronic device fabrication require
additional purification steps in order to produce electronic-grade silicon
(EGS). Electronic-grade silicon is also known as semiconductor-grade
silicon (SGS). In order for the purity levels to be acceptable for subsequent
crystal growth and device fabrication, EGS must have carbon and oxygen
impurity levels less than a few parts per million (ppm), and metal impurities
at the parts per billion (ppb) range or lower. [link] and [link] give typical

impurity concentrations in MGS and EGS, respectively. Besides the purity,
the production cost and the specifications must meet the industry desires.

Element Concentration Element Concentration
(ppm) (ppm)

aluminum 1000-4350 manganese 90-120

boron 40-60 molybdenum < 20

calcium 245-500 nickel 10-105

chromium 50-200 phosphorus 20-50

copper 15-45 titanium 140-300

iron 1550-6500 vanadium 50-250

magnesium 10-50 zirconium 20

Typical impurity concentrations found in metallurgical-grade silicon (MGS).

Elenite Concentration Element Concentration
(ppb) (ppb)
pene < 0.001 gold < 0.00001

antimony < 0.001 iron 0.1-1.0

boron < 0.1 nickel 0.1-0.5

carbon 100-1000 oxygen 100-400
chromium < 0.01 phosphorus < 0.3
cobalt 0.001 silver 0.001
copper 0.1 zinc <0.1

Typical impurity concentrations found in electronic-grade silicon (EGS).

Metallurgical-grade silicon (MGS)

The typical source material for commercial production of elemental silicon
is quartzite gravel; a relatively pure form of sand (SiO>). The first step in the
synthesis of silicon is the melting and reduction of the silica in a submerged-
electrode arc furnace. An example of which is shown schematically in
[link], along with the appropriate chemical reactions. A mixture of quartzite
gravel and carbon are heated to high temperatures (ca. 1800 °C) in the
furnace. The carbon bed consists of a mixture of coal, coke, and wood chips.
The latter providing the necessary porosity such that the gases created
during the reaction (SiO and CO) are able to flow through the bed.

quartzite, coal submerged electrode

coke, wood chips CO, SiO, H,O
naan ee ae
alias Si0+2C >"
form SiC
from SiO and C
melt SiO, KG co

te,

SiC + SiO, > Si+ SiO +CO

li id ili -
re discharge of MGSC—>

Schematic of submerged-electrode arc furnace
for the production of metallurgical-grade
silicon (MGS).

The overall reduction reaction of SiO, is expressed in [link], however, the
reaction sequence is more complex than this overall reaction implies, and
involves the formation of SiC and SiO intermediates. The initial reaction
between molten SiO» and C ([link]) takes place in the arc between adjacent
electrodes, where the local temperature can exceed 2000 °C. The SiO and
CO thus generated flow to cooler zones in the furnace where SiC is formed
({link]), or higher in the bed where they reform SiO, and C ([link]). The SiC
reacts with molten SiO> ({link]) producing the desired silicon along with
SiO and CO. The molten silicon formed is drawn-off from the furnace and

solidified.
Equation:

SiO,(liquid) + 2 C(solid) > Si(liquid) + 2 CO (gas)

Equation:

>1700 °C
Si0Q,+2C == Si0+CO
<1600 °C

Equation:
SiO0+2C > SiC +CO (1600 - 1700 °C)
Equation:

SiC + SiO, > Si+ SiO + CO

The as-produced MGS is approximately 98-99% pure, with the major
impurities being aluminum and iron ({link]), however, obtaining low levels
of boron impurities is of particular importance, because it is difficult to
remove and serves as a dopant for silicon. The drawbacks of the above
process are that it is energy and raw material intensive. It is estimated that
the production of one metric ton (1,000 kg) of MGS requires 2500-2700 kg
quartzite, 600 kg charcoal, 600-700 kg coal or coke, 300-500 kg wood chips,
and 500,000 kWh of electric power. Currently, approximately 500,000
metric tons of MGS are produced per year, worldwide. Most of the
production (ca. 70%) is used for metallurgical applications (e.g., aluminum-
silicon alloys are commonly used for automotive engine blocks) from
whence its name is derived. Applications in a variety of chemical products
such as silicone resins account for about 30%, and only 1% or less of the
total production of MGS is used in the manufacturing of high-purity EGS
for the electronics industry. The current worldwide consumption of EGS is
approximately 5 x 10° kg per year.

Electronic-grade silicon (EGS)

Electronic-grade silicon (EGS) is a polycrystalline material of exceptionally
high purity and is the raw material for the growth of single-crystal silicon.
EGS is one of the purest materials commonly available, see [link]. The
formation of EGS from MGS is accomplished through chemical purification

processes. The basic concept of which involves the conversion of MGS to a
volatile silicon compound, which is purified by distillation, and
subsequently decomposed to re-form elemental silicon of higher purity (i.e.,
EGS). Irrespective of the purification route employed, the first step is
physical pulverization of MGS followed by its conversion to the volatile
silicon compounds.

A number of compounds, such as monosilane (SiH,), dichlorosilane
(SiH»Cl)), trichlorosilane (SiHCl3), and silicon tetrachloride (SiCl,), have
been considered as chemical intermediates. Among these, SiHCl3 has been
used predominantly as the intermediate compound for subsequent EGS
formation, although SiH, is used to a lesser extent. Silicon tetrachloride and
its lower chlorinated derivatives are used for the chemical vapor deposition
(CVD) growth of Si and SiO>. The boiling points of silane and its
chlorinated products ([link]) are such that they are conveniently separated
from each other by fractional distillation.

Compound Boiling point (°C)
SiH, -112.3

SiH3Cl -30.4

SiH>Cl> 8.3

SiHCl3 31.5

SiCl, 57.6

Boiling points of silane and chlorosilanes at 760 mmHg (1 atmosphere).

The reasons for the predominant use of SiHCl3 in the synthesis of EGS are
as follows:

1. SiHCl; can be easily formed by the reaction of anhydrous hydrogen
chloride with MGS at reasonably low temperatures (200 - 400 °C);

2. it is liquid at room temperature so that purification can be
accomplished using standard distillation techniques;

3. it is easily handled and if dry can be stored in carbon steel tanks;

4. its liquid is easily vaporized and, when mixed with hydrogen it can be
transported in steel lines without corrosion;

5. it can be reduced at atmospheric pressure in the presence of hydrogen;
6. its deposition can take place on heated silicon, thus eliminating contact
with any foreign surfaces that may contaminate the resulting silicon;

and
7. it reacts at lower temperatures (1000 - 1200 °C) and at faster rates than
does SiCly.

Chlorosilane (Seimens) process

Trichlorosilane is synthesized by heating powdered MGS with anhydrous
hydrogen chloride (HCl) at around 300 °C in a fluidized-bed reactor, [link].
Equation:

ca. 300 °C
Si(solid) + 3 HCl(gas) == SiHCI,(vapor) + H, (gas)
>900 °C

Since the reaction is actually an equilibrium and the formation of SiHCl3
highly exothermic, efficient removal of generated heat is essential to assure
a maximum yield of SiHCl3. While the stoichiometric reaction is that shown
in Eq. 5, a mixture of chlorinated silanes is actually prepared which must be
separated by fractional distillation, along with the chlorides of any
impurities. In particular iron, aluminum, and boron are removed as FeCl;
(b.p. = 316 °C), AICI, (m.p. = 190 °C subl.), and BCI (b.p. = 12.65 °C),
respectively. Fractional distillation of SiHCl3 from these impurity halides
result in greatly increased purity with a concentration of electrically active
impurities of less than 1 ppb.

EGS is prepared from purified SiHCl3 in a chemical vapor deposition
(CVD) process similar to the epitaxial growth of Si. The high-purity SiHCl3
is vaporized, diluted with high-purity hydrogen, and introduced into the
Seimens deposition reactor, shown schematically in [link]. Within the
reactor, thin silicon rods called slim rods (ca. 4 mm diameter) are supported
by graphite electrodes. Resistance heating of the slim rods causes the
decomposition of the SiHCl3 to yield silicon, as described by the reverse
reaction shown in Eq. 5.

<—— reaction chamber

Si-bridge

Si-slim rod

Schematic representation of a Seimens
deposition reactor.

The shift in the equilibrium from forming SiHCl; from Si at low
temperature, to forming Si from SiHC]l; at high temperature is as a
consequence of the temperature dependence ({link]) of the equilibrium
constant ([link], where p = partial pressure) for [link]. Since the formation of
SiHCl3 is exothermic, i.e., AH < 0, an increase in the temperature causes the
partial pressure of SiHCls to decrease. Thus, the Siemens process is
typically run at ca. 1100 °C, while the reverse fluidized bed process is
carried out at 300 °C.

Equation:

InK, = -AH
RT

Equation:

Psincl, °H,

PHC

The slim rods act as a nucleation point for the deposition of silicon, and the
resulting polycrystalline rod consists of columnar grains of silicon
(polysilicon) grown perpendicular to the rod axis. Growth occurs at less than
1 mm per hour, and after deposition for 200 to 300 hours high-purity (EGS)
polysilicon rods of 150-200 mm in diameter are produced. For subsequent
float-zone refining the polysilicon EGS rods are cut into long cylindrical
rods. Alternatively, the as-formed polysilicon rods are broken into chunks
for single crystal growth processes, for example Czochralski melt growth.

In addition to the formation of silicon, the HCl] coproduct reacts with the
SiHCl3 reactant to form silicon tetrachloride (SiCl4) and hydrogen as major
byproducts of the process, [link]. This reaction represents a major
disadvantage with the Seimens process: poor efficiency of silicon and
chlorine consumption. Typically, only 30% of the silicon introduced into
CVD reactor is converted into high-purity polysilicon.

Equation:

HCl + SiHCl, > SiC, +H,

In order to improve efficiency the HCl, SiCl,, H», and unreacted SiHCl3 are
separated and recovered for recycling. [link] illustrates the entire
chlorosilane process starting with MGS and including the recycling of the
reaction byproducts to achieve high overall process efficiency. As a
consequence, the production cost of high-purity EGS depends on the
commercial usefulness of the byproduct, SiCl,. Additional disadvantages of
the Seimens process are derived from its relatively small batch size, slow

growth rate, and high power consumption. These issues have lead to the
investigation of alternative cost efficient routes to EGS.

Si(MGS) HC] SiC, HCl

hydrochlorination chlorosilane| Hz,
of Si (MGS) recovery
fluidized bed

reactor
SiHCl,

hydrogen
and HCl
revovery

SiHCl,
(SIH, Cly.x)

H2

SiHCls
vaporization H>
and chemical vapor
deposition

SiHCls

distillation

low boiling SiCly
impurities
Si (EGS)

Schematic representation of the reaction pathways
for the formation of EGS using the chlorosilane
process.

Silane process

An alternative process for the production of EGS that has begun to receive
commercial attention is the pyrolysis of silane (SiH). The advantages of
producing EGS from SiH, instead of SiHCls are potentially lower costs
associated with lower reaction temperatures, and less harmful byproducts.
Silane decomposes < 900 °C to give silicon and hydrogen, [Link].
Equation:

SiH,(vapor) > Si(solid) + 2 H, (gas)

Silane may be prepared by a number of routes, each having advantages with
respect to purity and production cost. The simplest process involves the
direct reaction of MGS powders with magnesium at 500 °C in a hydrogen
atmosphere, to form magnesium silicide (Mg»Si). The magnesium silicide is
then reacted with ammonium chloride in liquid ammonia below 0 °C, [link].
Equation:

Mg,Si+4NH,Cl > SiH, +2 MgCl, +5 NH,

This process is ideally suited to the removal of boron impurities (a p-type
dopant in Si), because the diborane (B>H¢) produced during the reaction
forms the Lewis acid-base complex, H3B(NH3), whose volatility is
sufficiently lower than SiHy, allowing for the purification of the latter. It is
possible to prepare EGS with a boron content of < 20 ppt using SiH,
synthesized in this manner. However, phosphorus (another dopant) in the
form of PH3 may be present as a contaminant requiring subsequent
purification of the SiHy.

Alternative routes to SiH, involve the chemical reduction of SiCl, by either
lithium hydride ([link]), lithium aluminum hydride ([link]), or via
hydrogenation in the presence of elemental silicon ([link] - [link]). The
hydride reduction reactions may be carried-out on relatively large scales (ca.
50 kg), but only batch processes. In contrast, Union Carbide has adapted the
hydrogenation to a continuous process, involving disproportionation
reactions of chlorosilanes ([link] - [link]) and the fractional distillation of
silane ({link]).

Equation:

SiCl,+4LiH > SiH, +4 LiCl
Equation:
SiCl, +4 LiAIH, > SiH,+LiCl + AICI,

Equation:

SiCl, +2 H, + Si(98%) > 4 SiHCI,
Equation:

2 SiHCI, > SiH,Cl, + SiCl,
Equation:

3 SiH,Cl, > SiH,CI +2 SiHCI,
Equation:

2 SiH,Cl > SiH, + SiH,Cl,

Pyrolysis of silane on resistively heated polysilicon filaments at 700-800 °C
yields polycrystalline EGS. As noted above, the EGS formed has
remarkably low boron impurities compared with material prepared from
trichlorosilane. Moreover, the resulting EGS is less contaminated with
transition metals from the reactor container because SiH, decomposition
does not cause as much of a corrosion problem as is observed with halide
precursor compounds.

Granular polysilicon deposition

Both the chlorosilane (Seimens) and silane processes result in the formation
of rods of EGS. However, there has been increased interest in the formation
of granular polycrystalline EGS. This process was developed in 1980’s, and
relies on the decomposition of SiH, in a fluidized-bed deposition reactor to

produce free-flowing granular polysilicon.

Tiny silicon particles are fluidized in a SiH4/Hp flow, and act as seed crystal
onto which polysilicon deposits to form free-flowing spherical particles. The
size distribution of the particles thus formed is over the range from 0.1 to 1.5
mm in diameter with an average particle size of 0.7 mm. The fluidized-bed

seed particles are originally made by grinding EGS in a ball (or hammer)
mill and leaching the product with acid, hydrogen peroxide, and water. This
process is time-consuming and costly, and tended to introduce undesirable
impurities from the metal grinders. In a new method, large EGS particles are
fired at each other by a high-speed stream of inert gas and the collision
breaks them down into particles of suitable size for a fluidized bed. This
process has the main advantage that it introduces no foreign materials and
requires no leaching or other post purification.

The fluidized-bed reactors are much more efficient than traditional rod
reactors as a consequence of the greater surface area available during CVD
growth of silicon. It has been suggested that fluidized-bed reactors require
‘7, to /19 the energy, and half the capital cost of the traditional process. The
quality of fluidized-bed polysilicon has proven to be equivalent to
polysilicon produced by the conventional methods. Moreover, granular EGS
in a free-flowing form, and with high bulk density, enables crystal growers
to obtain the high, reproducible production yields out of each crystal growth
run. For example, in the Czochralski crystal growth process, crucibles can
be quickly and easily filled to uniform loading with granular EGS, which
typically exceed those of randomly stacked polysilicon chunks produced by
the Siemens silane process.

Zone refining

The technique of zone refining is used to purify solid materials and is
commonly employed in metallurgical refining. In the case of silicon may be
used to obtain the desired ultimate purity of EGS, which has already been
purified by chemical processes. Zone refining was invented by Pfann, and
makes use of the fact that the equilibrium solubility of any impurity (e.g.,
Al) is different in the solid and liquid phases of a material (e.g., Si). For the
dilute solutions, as is observed in EGS silicon, an equilibrium segregation
coefficient (kp) is defined by kg = C./C), where C, and C; are the equilibrium
concentrations of the impurity in the solid and liquid near the interface,
respectively.

If kp is less than 1 then the impurities are left in the melt as the molten zone
is moved along the material. In a practical sense a molten zone is established
in a solid rod. The zone is then moved along the rod from left to right. If k <
1 then the frozen part left on the trailing edge of the moving molten zone
will be purer than the material that melts in on the right-side leading edge of
the moving molten zone. Consequently the solid to the left of the molten
zone is purer than the solid on the right. At the completion of the first pass
the impurities become concentrated to the right of the solid sample.
Repetition of the process allows for purification to exceptionally high levels.
[link]. lists the equilibrium segregation coefficients for common impurity
and dopant elements in silicon; it should be noted that they are all less than
1.

Element ko Element ko
aluminum 0.002 iron 8x 10°
boron 0.8 oxygen 0.25
carbon 0.07 phosphorus 0.35
copper 4x 10° antimony 0.023

Segregation coefficients for common impurity and dopant elements in
silicon.

Gallium arsenide

In contrast to electronic grade silicon (EGS), whose use is a minor fraction
of the global production of elemental silicon, gallium arsenide (GaAs) is
produced exclusively for use in the semiconductor industry. However,
arsenic and its compounds have significant commercial applications. The

main use of elemental arsenic is in alloys of Pb, and to a lesser extent Cu,
while arsenic compounds are widely used in pesticides and wood
preservatives and the production of bottle glass. Thus, the electronics
industry represents a minor user of arsenic. In contrast, although gallium has
minor uses as a high-temperature liquid seal, manometric fluids and heat
transfer media, and for low temperature solders, its main use is in
semiconductor technology.

Isolation and purification of gallium metal

At 19 ppm gallium (L. Gallia, France) is about as abundant as nitrogen,
lithium and lead; it is twice as abundant as boron (9 ppm), but is more
difficult to extract due to the lack of any major gallium-containing ore.
Gallium always occurs in association either with zinc or germanium, its
neighbors in the periodic table, or with aluminum in the same group. Thus,
the highest concentrations (0.1-1%) are in the rare mineral germanite (a
complex sulfide of Zn, Cu, Ge, and As), while concentrations in sphalerite
(ZnS), diaspore [AlO(OH)], bauxite, or coal, are a hundred-fold less.
Industrially, gallium was originally recovered from the flue dust emitted
during sulfide roasting or coal burning (up to 1.5% Ga), however, it is now
obtained as side product of vast aluminum industry and in particular from
the Bayer process for obtaining alumina from bauxite.

The Bayer process involves dissolution of bauxite, AlIO,OH3_>,, in aqueous
NaOH, separation of insoluble impurities, partial precipitation of the
trihydrate, Al(OH)3, and calcination at 1,200 °C. During processing the
alkaline solution is gradually enriched in gallium from an initial weight ratio
Ga/Al of about 1/5000 to about 1/300. Electrolysis of these extracts with a
Hg cathode results in further concentration, and the solution of sodium
gallate thus formed is then electrolyzed with a stainless steel cathode to give
Ga metal. Since bauxite contains 0.003-0.01% gallium, complete recovery
would yield some 500-1000 tons per annum, however present consumption
is only 0.1% of this about 10 tons per annum.

A typical analysis of the 98-99% pure gallium obtained as a side product
from the Bayer process is shown in [link]. This material is further purified to

99.99% by chemical treatment with acids and O> at high temperatures
followed by crystallization. This chemical process results in the reduction of
the majority of metal impurities at the ppm level, see [link]. Purification to
seven nines 99.9999% is possible through zone refining, however, since the
equilibrium distribution coefficient of the residual impurities kp ~ 1,
multiple passes are required, typically > 500. The low melting point of
gallium ensures that contamination from the container wall (which is
significant in silicon zone refining) is minimized. In order to facilitate the
multiple zone refining in a suitable time, a simple modification of zone
refining is employed shown in [link]. The gallium is contained in a plastic
tube wrapped around a rotating cylinder that is half immersed in a cooling
bath. A heater is positioned above the gallium plastic coil. Thus, establishing
a series of molten zones that pass upon rotation of the drum by one helical
segment per revolution. In this manner, 500 passes may be made in
relatively short time periods. The typical impurity levels of gallium zone
refined in this manner are given in [link].

Element

aluminum
calcium
copper
iron

lead

magnesium

Bayer
process
(ppm)
100-1,000
10-100
100-1,000
100-1,000
< 2000

10-100

After acid/base
leaching (ppm)

7
not detected
2
7

30

500 zone
Passes

(ppm)

<1

not detected
<1

|

not detected

not detected

mercury

nickel

silicon

tin

titanium

zinc

10-100
10-100
10-100
10-100
10-100

30,000

not detected not detected

not detected not detected
x1 not detected
x1] not detected
1 <1

x] not detected

Typical analysis of gallium obtained as a side product from the Bayer

process.

heater .
gallium
contained ina

rotating drum

plastic tube

Schematic representation of a zone refining

apparatus.

Isolation and purification of elemental arsenic

Elemental arsenic (L. arsenicum, yellow orpiment) exists in two forms:
yellow (cubic, As,) and gray or metallic (rhombohedral). At a natural
abundance of 1.8 ppm arsenic is relatively rare, however, this is offset by its
presence in a number of common minerals and the relative ease of isolation.

Arsenic containing minerals are grouped into three main classes: the sulfides
realgar (As,4S,) and orpiment (As>S3), the oxide arsenolite (As,O3), and the
arsenides and sulfaresenides of the iron, cobalt, and nickel. Minerals in this
latter class include: loellinginite (FeAs>), safforlite (CoAs), niccolite (NiAs),
rammelsbergite (NiAs>), ansenopyrite or mispickel (FeAsS), cobaltite
(CoAsS), enargite (Cu3AsS,), gerdsorfite (NiAsS), and the quarturnary
sulfide glaucodot [(Co,Fe)AsS]. [link] shows the typical impurities in
arsenopyrite.

Element Concentration Element Concentration
(ppm) (ppm)

silver 90 nickel < 3,000

gold 8 lead 50

cobalt 30,000 platinum 0.4

copper 200 rhenium 50

germanium 30 selenium 50

manganese 3,000 vanadium 300

molybdenum 60 zinc 400

Typical impurities in arsenopyrite.

Arsenic is obtained commercially by smelting either FeAs» or FeAsS at 650-
700 °C in the absence of air and condensing the sublimed element (Ts,4 =
613 °C), [link].

Equation:

650-700 °C <613 °C
FeAsS > FeS+As(vapor) > As(solid)

The arsenic thus obtained is combined with lead and then sublimed (T,,, =
614 °C) which binds any sulfur impurities more strongly than arsenic. Any
residual arsenic that remains trapped in the iron sulfide is separated by
forming the oxide (As»O3) by roasting the sulfide in air. The oxide is
sublimed into the flue system during roasting from where it is collected and
reduced with charcoal at 700-800 °C to give elemental arsenic.
Semiconductor grade arsenic (> 99.9999%) is formed by zone refining.

Synthesis and purification of gallium arsenide.

Gallium arsenide can be prepared by the direct reaction of the elements,
[link]. However, while conceptually simple the synthesis of GaAs is
complicated by the different vapor pressures of the reagents and the highly
exothermic nature of the reaction. Furthermore, since the synthesis of GaAs
at atmospheric pressure is accompanied by its simultaneous decomposes due
to the loss by sublimation, of arsenic, the synthesis must be carried out
under an overpressure of arsenic in order to maintain a stoichiometric
composition of the synthesized GaAs.

Equation:

>1240 °C
Ga(liquid) + As(vapor) > GaAs(solid)

In order to overcome the problems associated with arsenic loss, the reaction
is usually carried out in a sealed reaction tube. However, if a stoichiometric
quantity of arsenic is used in the reaction a constant temperature of 1238 °C
must be employed in order to maintain the desired arsenic overpressure of 1
atm. Practically, it is easier to use a large excess of arsenic heated to a lower
temperature. In this situation the pressure in the tube is approximately equal
to the equilibrium vapor pressure of the volatile component (arsenic) at the
lower temperature. Thus, an over pressure of 1 atm arsenic may be

maintained if within a sealed tube elemental arsenic is heated to 600-620 °C
while the GaAs is maintained at 1240-1250 °C.

[link] shows the sealed tube configuration that is typically used for the
synthesis of GaAs. The tube is heated within a two-zone furnace. The boats
holding the reactants are usually made of quartz, however, graphite is also
used since the latter has a closer thermal expansion match to the GaAs
product. If higher purity is required then pyrolytic boron nitride (PBN) is
used. One of the boats is loaded with pure gallium the other with arsenic. A
plug of quartz wool may be placed between the boats to act as a diffuser.
The tube is then evacuated and sealed. Once brought to the correct reaction
temperatures ([link]), the arsenic vapor is transported to the gallium, and
they react to form GaAs in a controlled manner. [link] gives the typical
impurity concentrations found in polycrystalline GaAs.

arsenic gallium

VEZERY)

ITT) TTT

600 - 620 °C 1240 - 1260 °C

Schematic representation of a sealed tube

synthesis of GaAs.
Ficmeni Concentration Ficnient Concentration
(ppm) (ppm)

boron 0.1 silicon 0.02

carbon 0.7 phosphorus 0.1

nitrogen 0.1 sulfur 0.01
oxygen 0.5 chlorine 0.08
fluorine 0.2 nickel 0.04
magnesium 0.02 copper 0.01
aluminum 0.02 zinc 0.05

Impurity concentrations found in polycrystalline GaAs.

Polycrystalline GaAs, formed in from the direct reaction of the elements is
often used as the starting material for single crystal growth via Bridgeman or
Czochralski crystal growth. It is also possible to prepare single crystals of
GaAs directly from the elements using in-situ, or direct, compounding
within a high-pressure liquid encapsulated Czochralski (HPLEC) technique.

Bibliography

e K.G. Baraclough, K. G., in The Chemistry of the Semiconductor
Industry, Eds. S. J. Moss and A. Ledwith, Blackie and Sons, Glasgow,
Scotland (1987).

e L. D. Crossman and J. A. Baker, Semiconductor Silicon 1977,
Electrochem. Soc., Princeton, New Jersey (1977).

e M. Fleisher, in Economic Geology, 50th Aniv. Vol., The Economic
Geology Publishing Company, Lancaster, PA (1955).

e G. Hsu, N. Rohatgi, and J. Houseman, AIChE J., 1987, 33, 784.

e S.K. lya, R. N. Flagella, and F. S. Dipaolo, J. Electrochem. Soc., 1982,
129, 1531.

e J. Krauskopf, J.D. Meyer, B. Wiedemann, M. Waldschmidt, K. Bethge,
G. Wolf, and W. Schiiltze, 5th Conference on Semi-insulating II-V
Materials, Malmo, Sweden, 1988, Eds. G. Grossman and L. Ledebo,
Adam-Hilger, New York (1988).

J. R. McCormic, Conf. Rec. 14th IEEE Photovolt. Specialists Conf.,
San Diego, CA (1980).

J. R. McCormic, in Semiconductor Silicon 1981, Ed. H. R. Huff,
Electrochemical Society, Princeton, New Jersey (1981).

W. C. O’ Mara, Ed. Handbook of Semiconductor Silicon Technology,
Noyes Pub., New Jersey (1990).

W. G. Pfann, Zone Melting, John Wiley & Sons, New York, (1966).
F, Shimura, Semiconductor Silicon Crystal Technology, Academic
Press (1989).

Growth of Gallium Arsenide Crystals

Introduction

When considering the synthesis of Group 13-15 compounds for electronic
applications, the very nature of semiconductor behavior demands the use of
high purity single crystal materials. The polycrystalline materials
synthesized above are, therefore, of little use for 13-15 semiconductors but
may, however, serve as the starting material for melt grown single crystals.
For GaAs, undoubtedly the most important 13-15 (III - V) semiconductor,
melt grown single crystals are achieved by one of two techniques: the
Bridgman technique, and the Czochralski technique.

Bridgman growth

The Bridgman technique requires a two-zone furnace, of the type shown in
[link]. The left hand zone is maintained at a temperature of ca. 610 °C,
allowing sufficient overpressure of arsenic within the sealed system to
prevent arsenic loss from the gallium arsenide. The right hand side of the
furnace contains the polycrystalline GaAs raw material held at a
temperature just above its melting point (ca. 1240 °C). As the furnace
moves from left to right, the melt cools and solidifies. If a seed crystal is
placed at the left hand side of the melt (at a point where the temperature
gradient is such that only the end melts), a specific orientation of single
crystal may be propagated at the liquid-solid interface eventually to produce
a single crystal.

furnace zone 1 furnace zone 2

; seed GaAs
arsenic crystal charge

Direction of furnace travel

A schematic diagram of a Bridgman two-zone
furnace used for melt growths of single crystal
GaAs.

Czochralski growth

The Czochralski technique, which is the most commonly used technique in
industry, is shown in [link]. The process relies on the controlled withdrawal
of a seed crystal from a liquid melt. As the seed is lowered into the melt,
partial melting of the tip occurs creating the liquid solid interface required
for crystal growth. As the seed is withdrawn, solidification occurs and the
seed orientation is propagated into the grown material. The variable
parameters of rate of withdrawal and rotation rate can control crystal
diameter and purity. As shown in [link] the GaAs melt is capped by boron
trioxide (B03). The capping layer, which is inert to GaAs, prevents arsenic
loss when the pressure on the surface is above atmospheric pressure. The
growth of GaAs by this technique is thus termed liquid encapsulated
Czochralski (LEC) growth.

counter-
clockwise
rotation

seed crystal
fused ‘

silica
crucible
R.F. Coils

single crystal graphite
susceptor

\
° |. © tone
fo) pp cap
(@) Ee | iii tii
ce) oO
O 1@)
ce) te)
liquid melt

clockwise rotation

A schematic diagram of the Czochralski
technique as used for growth of GaAs single
crystal bond.

While the Bridgman technique is largely favored for GaAs growth, larger
diameter wafers can be obtained by the Czochralski method. Both of these
melt techniques produce materials heavily contaminated by the crucible,
making them suitable almost exclusively as substrate material. Another
disadvantage of these techniques is the production of defects in the material
caused by the melt process.

Bibliography

e W.G. Pfann, Zone Melting, John Wiley & Sons, New York (1966).

e R.E. Williams, Gallium Arsenide Processing Techniques. Artech
House (1984).

Ceramic Processing of Alumina

Introduction

While aluminum is the most abundant metal in the earth's crust (ca. 8%)
and aluminum compounds such as alum, K[AI(SO,)].12(H2O), were
known throughout the world in ancient times, it was not until the isolation
of aluminum in the late eighteenth century by the Danish scientist H. C.
Oersted that research into the chemistry of the Group 13 elements began in
earnest. Initially, metallic aluminum was isolated by the reduction of
aluminum trichloride with potassium or sodium; however, with the advent
of inexpensive electric power in the late 1800's, it became economically
feasible to extract the metal via the electrolyis of alumina (Al,O3) dissolved
in cryolite, Na3AlF¢, (the Hall-Heroult process). Today, alumina is prepared
by the Bayer process, in which the mineral bauxite (named for Les Baux,
France, where it was first discovered) is dissolved with aqueous hydroxides,
and the solution is filtered and treated with CO> to precipitate alumina.
With availability of both the mineral and cheap electric power being the
major considerations in the economical production of aluminum, it is not
surprising that the leading producers of aluminum are the United States,
Japan, Australia, Canada, and the former Soviet Union.

Aluminum oxides and hydroxides

The many forms of aluminum oxides and hydroxides are linked by complex
structural relationships. Bauxite has the formula Al,(OH)3.9, (0 < x < 1)
and is thus a mixture of Al5O3 (a-alumina), Al(OH) 3 (gibbsite), and
AlO(OH) (boehmite). The latter is an industrially important compound
which is used in the form of a gel as a pre-ceramic in the production of
fibers and coatings, and as a fire retarding agent in plastics.

Heating boehmite and diaspore to 450 °C causes dehydration to yield forms
of alumina which have structures related to their oxide-hydroxide
precursors. Thus, boehmite produces the low-temperature form y-alumina,
while heating diaspore will give a-alumina (corundum). y-alumina converts
to the hcp structure at 1100 °C. A third form of Al»O3 forms on the surface
of the clean aluminum metal. The thin, tough, transparent oxide layer is the

reason for much of the usefulness of aluminum. This oxide skin is rapidly
self-repairing because its heat of formation is so large (AH = -3351 kJ/mol).
Equation:

4Al +30, > 2A1,0,

Ternary and mixed-metal oxides

A further consequence of the stability of alumina is that most if not all of
the naturally occurring aluminum compounds are oxides. Indeed, many
precious gemstones are actually corundum doped with impurities.
Replacement of aluminum ions with trace amounts of transition-metal ions
transforms the formerly colorless mineral into ruby (red, Cr°*), sapphire
(blue, Fe**/3*, Ti**), or topaz (yellow, Fe**). The addition of stoichiometric
amounts of metal ions causes a shift from the a-Al,O3 hcp structure to the
other common oxide structures found in nature. Examples include the
perovskite structure for ABO3 type minerals (e.g., CeTiO7 or LaAlO3) and
the spinel structure for AB»O, minerals (e.g., beryl, BeAl Ox).

Aluminum oxide also forms ternary and mixed-metal oxide phases. Ternary
systems such as mullite (AlgSi,O;3), yttrium aluminum garnet (YAG,
Y3AI150 >), the B-aluminas (e.g., NaAl,;;0,7) and aluminates such as
hibonite (CaAl,»019) possessing B-alumina or magnetoplumbite-type
structures can offer advantages over those of the binary aluminum oxides.

Applications of these materials are found in areas such as engineering
composite materials, coatings, technical and electronic ceramics, and
catalysts. For example, mullite has exceptional high temperature shock
resistance and is widely used as an infrared-transparent window for high
temperature applications, as a substrate in multilayer electronic device
packaging, and in high temperature structural applications. Hibonite and
other hexaluminates with similar structures are being evaluated as
interfacial coatings for ceramic matrix composites due to their high thermal
stability and unique crystallographic structures. Furthermore, aluminum
oxides doped with an alkali, alkaline earth, rare earth, or transition metal are

of interest for their enhanced chemical and physical properties in
applications utilizing their unique optoelectronic properties.

Synthesis of aluminum oxide ceramics

In common with the majority of oxide ceramics, two primary synthetic
processes are employed for the production of aluminum oxide and mixed
metal oxide materials:

1. The traditional ceramic powder process.
2. The solution-gelation, or "sol-gel" process.

The environmental impact of alumina and alumina-based ceramics is in
general negligible; however, the same cannot be said for these methods of
preparation. As practiced commercially, both of the above processes can
have a significant detrimental environmental impact.

Traditional ceramic processing

Traditional ceramic processing involves three basic steps generally referred
to as powder-processing, shape-forming, and densification, often with a
final mechanical finishing step. Although several steps may be energy
intensive, the most direct environmental impact arises from the shape-
forming process where various binders, solvents, and other potentially toxic
agents are added to form and stabilize a solid ("green") body ({link]).

Volume

Function Composition (%)

Powder alumina (Al,O3) 27

Solvent 1,1,1-trichloroethane/ethanol 58
Deflocculant menhaden oil 1.8
Binder poly(vinyl butyrol) 4.4

poly(ethylene glycol)/octyl

phthalate oe

Plasticizer

Typical composition of alumina green body

The component chemicals are mixed to a slurry, cast, then dried and fired.
In addition to any innate health risk associated with the chemical processing
these agents are subsequently removed in gaseous form by direct
evaporation or pyrolysis. The replacement of chlorinated solvents such as
1,1,1-trichloroethylene (TCE) must be regarded as a high priority for
limiting environmental pollution. The United States Environmental
Protection Agency (EPA) included TCE on its 1991 list of 17 high-priority
toxic chemicals targeted for source reduction. The plasticizers, binders, and
alcohols used in the process present a number of potential environmental
impacts associated with the release of combustion products during firing of
the ceramics, and the need to recycle or discharge alcohols which, in the
case of discharge to waterways, may exert high biological oxygen demands
in the receiving communities. It would be desirable, therefore, to be able to
use aqueous processing; however, this has previously been unsuccessful due
to problems associated with batching, milling, and forming. Nevertheless,
with a suitable choice of binders, etc., aqueous processing is possible.
Unfortunately, in many cast-parts formed by green body processing the
liquid solvent alone consists of over 50 % of the initial volume, and while
this is not directly of an environmental concern, the resultant shrinkage
makes near net shape processing difficult.

Sol-gel

Whereas the traditional sintering process is used primarily for the
manufacture of dense parts, the solution-gelation (sol-gel) process has been

applied industrially primarily for the production of porous materials and
coatings.

Sol-gel involves a four stage process: dispersion, gelation, drying, and
firing. A stable liquid dispersion or sol of the colloidal ceramic precursor is
initially formed in a solvent with appropriate additives. By changing the
concentration (aging) or pH, the dispersion is "polymerized" to form a solid
dispersion or gel. The excess liquid is removed from this gel by drying and
the final ceramic is formed by firing the gel at higher temperatures.

The common sol-gel route to aluminum oxides employs aluminum
hydroxide or hydroxide-based material as the solid colloid, the second
phase being water and/or an organic solvent, however, the strong
interactions of the freshly precipitated alumina gels with ions from the
precursor solutions makes it difficult to prepare these gels in pure form. To
avoid this complication, alumina gels are also prepared from the hydrolysis
of aluminum alkoxides, Al(OR)3.

Equation:

AI(OR), + H,O/Ht > Al-gel

Equation:

A
Al-gel > ALO,

The exact composition of the gel in commercial systems is ordinarily
proprietary, however, a typical composition will include an aluminum
compound, a mineral acid, and a complexing agent to inhibit premature
precipitation of the gel, e.g., [link].

Function Composition

Boehmite precursor ASB [aluminum sec-butoxide, Al(OC4Ho)3]
Electrolyte HNO3 0.07 mole/mole ASB
Complexing agent glycerol ca. 10 wt.%

Typical composition of an alumina sol-gel for slipcast ceramics.

The principal environmental consequences arising from the sol-gel process
are those associated with the use of strong acids, plasticizers, binders,
solvents, and sec-butanol formed during the reaction. Depending on the
firing conditions, variable amounts of organic materials such as binders and
plasticizers may be released as combustion products. NO,’s may also be
produced in the off-gas from residual nitric acid or nitrate salts. Moreover,
acids and solvents must be recycled or disposed of. Energy consumption in
the process entails “upstream” environmental emissions associated with the
production of that energy.

Bibliography

e Advances in Ceramics, Eds. J. A. Mangels and G. L. Messing,
American Ceramic Society, Westville, OH, 1984, Vol. 9.

e Adkins, J. Am. Chem. Soc., 1922, 44, 2175.

e A.R. Barron, Comm. Inorg. Chem., 1993, 14, 123.

e M. K. Cinibulk, Ceram. Eng. Sci., Proc., 1994, 15, 721.

e F. A. Cotton and G. Wilkinson, Advanced Inorganic Chemistry, 5th
Ed., John Wiley and Sons, New York (1988).

e N. N. Greenwood and A. Earnshaw, Chemistry of the Elements,
Pergamon Press, Oxford (1984).

e P.H. Hsu and T. F. Bates, Mineral Mag., 1964, 33, 749.

e W. D. Kingery, H. K. Bowen, and D. R. Uhlmann, Introduction to
Ceramics, 2nd Ed. Wiley, New York (1976).

e H. Schneider, K. Okada, and J. Pask, Mullite and Mullite Ceramics,
Wiley (1994).

e R. V. Thomas, Systems Analysis and Water Quality Management,
McGraw-Hill, New York (1972).

e J.C. Williams, in Treatise on Materials Science and Technology, Ed. F.
F. Y. Wang, Academic Press, New York (1976).

Piezoelectric Materials Synthesis

This module was developed as part of the Rice University course CHEM-496: Chemistry of Electronic
Materials. This module was prepared with the assistance of Ilse Y. Guzman-Jimenez.

Introduction

Piezoelectricity is the generation of an electric moment by a change of stress applied to a solid. The word
piezoelectricity literally means “pressure electricity”; the prefix piezo is derived from the Greek word
piezein, “to press”. The piezoelectric effect was discovered in 1880 by the brothers Jacques and Pierre
Curie. Not only did they demonstrate the phenomenon, but they also established the criteria for its
existence in a given crystal. Of the thirty-two crystal classes, twenty-one are non-centrosymmetric (not
having a centre of symmetry), and of these, twenty exhibit direct piezoelectricity.

The first practical application of the piezoelectric effect was developed when ground quartz crystals were
placed between the plates of a tuning capacitor in order to stabilize oscillating circuits in radio transmitters
and receivers; however, the phenomenon of piezoelectricity was not well exploited until World War I,
when Langevin used piezoelectrically excited quartz plates to generate sounds waves in water for use in
submarine detection.

Piezoelectricity can also occur in polycrystalline or amorphous substances which have become anisotropic
by external agents. Synthetic piezoelectric materials became available near the end of World War II, with
the accidental discovery of the fact that materials like barium titanate and rare earth oxides become
piezoelectric when they are polarized electrically. During the postwar years, when germanium and silicon
were revolutionizing the electronics industry, piezoceramics appeared for a while to be joining the
revolution, but the limited availability of materials and components, made the piezoelectric phenomenon
failed to lead mature applications during the 1950s. It is only now that a variety of piezoelectric materials
are being synthesized and optimized. As a consequence piezoelectric-based devices are undergoing a
revolutionary development, specially for medicine and aerospace applications.

Piezoelectric ceramics

Most piezoelectric transducers are made up of ceramic materials for a broad range of electromechanical
conversion tasks as transmitters, ranging from buzzers in alarm clocks to sonars, and as receivers, ranging
from ultra high frequency (UHF) filters to hydrophones.

Most of the piezoelectric materials in usage are from the lead zirconate titanate (PZT) family, because of
their excellent piezoelectric parameters, thermal stability, and dielectric properties. Additionally the
properties of this family can be modified by changing the zirconium to titanium ratio or by addition of both
metallic and non-metallic elements. PZT (PbZr,_,TixO3) ceramics and their solid solutions with several
complex perovskite oxides have been studied; among the various complex oxide materials, niobates have
attracted special attention. Ternary ceramic materials, lead metaniobate, as well as, barium and modified
lead titanates complete the list of piezoceramic materials.

Selective parameters for piezoceramic materials are given in [link], where Q,, is the mechanical quality
factor, T, is the Curie point, d3; is the the transverse charge coefficient, and kp, k;, and k3, are the
electromechanical coupling factors for planar, thickness, and transversal mode respectively.

Material PZT Lead PSZNT PZT, PSN- TsTS- PZT,

property modified metaniobate 31/40/29 x= PLT 42-1 x=
0.5 50/50 0.48

Qn 350 40 222 74 41 887

T. (°C) 290 462 369 152 355

CN 50

kp 0.5 60 0.428 30.7 46.5

k 0.32 0.438 -

k31 0.21 0.263 17.9

Selective parameters for illustrative piezoceramic materials.

Recently, sol-gel processing has been used to prepare ceramics, making possible the preparation of
materials that are difficult to obtain by conventional methods. Both, inorganic and organic precursor have
been reported. Additionally, new techniques for the production of ceramic fibers have been developed.
Better processing and geometrical and microestructural control are the main goals in the production of
fibers.

The latest development in piezoceramic fibers is the modification of the viscous-suspension-spinning
process (VSSP) for the production of continuos piezoelectric ceramic fibers for smart materials and active
control devices, such as transducers, sensor/actuators and structural-control devices. The VSSP utilizes
conventional synthesized ceramic powders and cellulose, as the fugitive carrier, to produce green ceramic
fiber at a reasonable cost. [link] shows the schematic representation of the VSSP.

Regeneration

bath
Ceramic ——— Wash drum
dispersion os ————— Dryer Take-up

Mix Filter and de-air 1S) <= 5 { drum reel

} 4) | an

: list — hy fre oe Gas
—_e— SF =
Metering pump Spin bath Finish bath

The viscous-suspension-spinning process (VSSP) for the production
of continuous piezoceramic fiber.

Synthesis of reactive PZT precursor powder by the oxalate coprecipitation technique has also been
developed. The precursor transforms to phase pure PZT at or above 850 °C the PZT obtained by this
technique showed a Curie temperature of 355 °C. The advantages of the coprecipitation technique are the
lack of moisture sensitive and special handling precursors.

Although new materials have been investigated with the purpose of create replacements for ceramics, there
has been a great improvement in their properties and, current research is focused in the development of

new techniques for both synthesis and processing.

Piezoelectric single crystals.

The recent progress of the electronic technology requires new piezoelectric crystals with a high thermal
stability and large electromechanical coupling factors. Single-crystal materials have been considered as
replacements for polycrystalline ceramics. Ideally single-crystals of lead zirconate titanate (PZT) itself
would be the main choice as it is the most prevailing piezoelectric material, but it is difficult to grow large
single crystals. On the other hand, the fact that single-crystals offer many advantages over polycrystalline
systems has been recognized. Materials such as lithium niobate present essentially no aging, no mechanical
creep and excellent performance in high temperature conditions.

New piezoelectric single crystals grown by conventional RF-heating Czochralski (CZ) technique have
been synthesized. High purity starting materials, mainly oxides powders, and Ar atmosphere are required.
La3GasSiOj4, Lag3Nbg 5Gas.5O14 and La3Tag.5Gas.5O1,4 single crystals have been grown by using this
method. However, the CZ technique can be applied only to materials that can be synthesized by ordinary
solid-state reaction and can undergo the pulling method.

BaBe)Si,O7 (barylite) has been known as material with a strong piezoelectricity, however, it can not be
obtained by solid-state reaction and CZ technique therefore is not applicable. As an alternative for
piezoelectric crystals growth hydrothermal synthesis has been developed. [link] shows the experimental
apparatus for the growth of barylite. Eventhough, crystals can be obtained using this technique, high
pressure (500 - 1000 bar) and a solvent for the raw materials are required.

Experimental
apparatus for the
hydrothermal
synthesis of
barylite. H = heater,
F = furnace, S =
specimen vessel, G
= growth capsule, P
= pressure gauge,
and T =
thermocouples.

Adapted from M.
Maeda, T. Uehara,
H. Sato and T.
Ikeda, Jpn. J. Appl.
Phys., 1991, 30,
2240.

While the piezoceramics dominate the single crystal materials in usage, single crystals piezoelectrics
continue to make important contributions both in price-conscious consumer market and in performance-
driven defense applications. Areas such as frequency stabilized oscillators, surface acoustic wave devices
and filters with a wide pass band, are still dominated by single crystals.

Piezoelectric thin films

Recently, there has been great interest in the deposition of piezoelectric thin films, mainly for
microelectronical systems (MEMS) applications; where the goal is to integrate sensors and actuators based
on PZT films with Si semiconductor-based signal processing; and for surface acoustic wave (SAW)
devices; where the goal is to achieve higher electromechanical coupling coefficient and temperature
stability. Piezoelectrical microcantilevers, microactuators, resonators and SAW devices using thin films
have been reported.

Several methods have been investigated for PZT thin films. In the metallo-organic thin film deposition,
alkoxides are stirred during long periods of time (up to 18 hours). After pyrolisis, PZT amorphous films are
formed and then calcination between 400 — 600 °C for 80 hours leads to PZT crystallization (perovskita
phase) by a consecutive phase transformation process, which involves a transitional pyrochlore phase.

A hybrid metallorganic decomposition (MOD) route has also been developed to prepare PZT thin films.
Lead and titanium acetates and, zirconium acetylacetonate are used. The ferroelectric piezoelectric and
dielectric properties indicate that the MOD route provides PZT films of good quality and comparable to
literature values. In addition to being simple, MOD has several advantages which include: homogeneity at
molecular level and ease composition control.

Metalorganic chemical vapor deposition (MOCVD) has been applied to PZT thin films deposition also. It
has been proved that excellent quality PZT films can be grown by using MOCVD, but just recently the
control of microstructure the deposition by varying the temperature, Zr to Ti ratio and precursors flow has
been studied. Recent progress in PZT films deposition has led to lower temperature growth and it is
expected that by lowering the deposition temperature better electrical properties can be achieved.
Additionally, novel techniques such as KrF excimer laser ablation and, ion and photo-assisted depositions,
have also been used for PZT films synthesis.

On the other hand, a single process to deposit PZT thin film by a hydrothermal method has been reported
recently. Since the sol-gel method, sputtering and chemical vapor deposition techniques are useful only for
making flat materials, the hydrothermal method offers the advantage of making curved shaped materials.
The hydrothermal method utilizes the chemical reaction between titanium and ions melted in solution. A
PZT thin film has been successfully deposited directly on a titanium substrate and the optimum ion ratio in
the solution is being investigated to improve the piezoelectric effect.

Among the current reported piezoelectric materials, the Pb(Ni,/3Nb2/3)9.2Zro,4Tig. 403 (PNNZT, 2/4/4)
ferroelectric ceramic has piezoelectric properties that are about 60 and 3 times larger than the reported
values for ZnO and PZT. A sol-gel technique has been developed for the deposition of a novel piezoelectric

PNNZT thin film. A 2-methoxyethanol based process is used. In this process precursors are heated at lower
temperature than the boiling point of the solvent, to distill off water. Then prior high temperature
annealing, addition of excess Pb precursor in the precursor solution is required to compensate the lead loss.
The pure perovskite phase is then obtained at 600 °C, after annealing.

Thin films of zinc oxide (ZnO), a piezoelectric material and n-type wide-bandgap semiconductor, have
been deposited. ZnO films are currently used in SAW devices and in electro-optic modulators. ZnO thin
films have been grown by chemical vapor deposition and both d.c. and r.f. sputtering techniques. Recently,
optimization of ZnO films by r.f. magnetron sputtering has been developed. However, homogeneity is one
of the main problems when using this technique, since films grown by this optimized method, showed two
regions with different piezoelectric properties.

DC magnetron sputtering is other technique for piezoelectric thin film growth, recently aluminum nitride, a
promising material for use in thin-film bulk acoustic wave resonators for applications in RF bandpass
filters, has been grown by this method. The best quality films are obtained on Si substrates. In order to
achieve the highest resonator coupling, the AIN must be grown directly on the electrodes. The main
problem in the AIN growth is the oxygen contamination, which leads to the formation of native oxide on
the Al surface, preventing crystalline growth of AIN.

Piezoelectric polymers

The discovery of piezoelectricity in polymeric materials such as polyvinylidene difluoride (PVF), was
considered as an indication of a renaissance in piezoelectricity. Intensive research was focused in the
synthesis and functionalization of polymers. A potential piezoelectric polymer has to contain a high
concentration of dipoles and also be mechanically strong and film-forming. The degree of crystallinity and
the morphology of the crystalline material have profound effects on the mechanical behavior of polymers.
Additionally, in order to induce a piezoelectric response in amorphous systems the polymer is poled by
application of a strong electric field at elevated temperature sufficient to allow mobility of the molecular
dipoles in the polymer. Recent approaches have been focused in the development of cyano-containing
polymers, due to the fact that cyano polymers could have many dipoles which can be aligned in the same
direction.

Phase transfer catalyzed reaction has been used for piezoelectric polymer preparation from malonitrile,
however this method leads to low molecular weight, and low yield of impure vinylidene cyanide units
containing material. The use of solid K>CO3 and acetonitrile without added phase transfer catalyst shows
excellent yields for polyester possessing backbone gem-dinitriles and for polyamide synthesis. The
polyester and polyamide obtained contained a dinitrile group net dipole which can be align in the same
direction as the carbonyl groups.

The pursuit for better piezoelectric polymers has led to molecular modeling which indicates that one cyano
substituent should be almost as effective as two geminal cyano substituents, opening a new area of
potential materials having an acrylonitrile group as the basic building block. However, polyacrilonitrile
itself is not suitable because it forms a helix. Thus acrylonitrile copolymers have been investigated.

Most of the piezoelectric polymers available are still synthesized by conventional methods such as
polycondensation and radical polymerization. Therefore piezoelectric polymer synthesis has the same
problems as the commercial polymer preparation, such as controlling the degree of polymerization and
crystallinity.

A novel technique of vapor deposition polymerization has been reported as an alternative method to
copolymeric thin films. Aliphatic polyurea 9 was synthesized by evaporating monomers of 1,9-
diaminononano and 1,9-diisocyanatononano onto glass substrate in vacuum. Deposition rates were
improved at temperatures below 0 °C. After poling treatment films showed fairly large piezoelectric

activities. Additionally, a completely novel approach to piezoelectric polymers has been presented. This
approach, consists in the synthesis of ordered piezoelectric polymer networks via crosslinking of liquid-
crystalline monomers. The main goal in this approach is to achieve a polymer network which combines the
long term stability of piezoelectric single-crystals with the ease of processability and fabrication of
conventional polymers. [link] shows the schematic representation of this approach.

a. F ae.
S SSSSSSSE Orient layers LEITIITIPPR hv
ANAANAAASASSS E field across plates LAGI p> Polymerize in-situ
Chiral S,* monomers Poled S.* monomers Crosslinked network
Local helical symmetry Uniform bulk C, symmetry Bulk C, symmetry
Fluid Fluid Non-centrosymmetric solid

Scheme of a ordered piezoelectric networks via a liquid-crystalline
monomer strategy. Adapted from D. L. Gin and B. C. Baxter,
Polymer Preprints, 1996, 38, 211.

Piezoelectric polymers are becoming increasingly important commercially because of their easier
processability, lower cost, and higher impact resistance than ceramics, but the lack of high temperature
stability and the absence of a solid understanding of the molecular level basis for the electrical properties
are limitations. The requirements for strong piezoelectricity in a polymer are: the polymer chain has a
larger resultant dipole moment normal to the chain axis; polymer crystallizes into a polar crystal with the
polar axis perpendicular to the chain axis, has a high crystallinity and finally the polymer polar axis aligns
easily in the thickness direction during poling.

Piezoelectric composites

Piezocomposites have been obtained by the combination of piezoelectric ceramics and polymers, the
resulting material posses both the high piezoelectric properties of ceramics and the processability of
polymers. 1-3 type piezocomposites have found wide applications as medical and industrial ultrasonic
transducers.

The current method for piezocomposite production is the dice-and-fill technique, which consists in cutting
two sets of grooves in a block of piezoceramic at right angles each other, then a polymer is cast into these
grooves and the solid ceramic base is ground off. Polishing and poling are the following steps in order to
achieve the final thickness and properties. This method is expensive, time consuming and size limited.

As an alternative for the dice-and-fill technique, continuos green fibers obtained by the modified viscous-
suspension-spinning process, can be bundled into a cottonball-like shape, then burned and sintered. The
sintered bundle impregnated with epoxy resin can be sliced into discs and then polarized. Recent results
have yielded 1-3 type composites with excellent piezoelectric properties.

On the other hand, an innovative process has been developed for Srj(Nbp.5Tag.5)207/PVDF composites, in
this new fabrication method, appropriate amounts of oxides are mixed, pressed and sintered. The porous
resulting material is subsequently infiltrated with PVDF solution and then poled. This new method for
composites preparation is simple and offers a lead-free alternative smart material.

Another kind of piezocomposites can be achieved by spinning films of piezoceramic onto metal alloys,
such as TiNi. The resulting materials is a hybrid composite that can utilize the different active and adaptive

properties of the individual bulk materials. Due to the shape memory nature of TiNi, a possible application
for this new heterostructures could be smart active damping of mechanical vibrations. DC sputtering and
spin coating are the techniques necessary for the smart thin film TiNi/piezoelectric heterostructures
fabrication. However, eventhough the films had a fine grain structure and high mechanical qualities, the
ferroelectric properties were poor compared to literature values.

In the future, the properties of piezocomposites will be tailored, by varying the ceramic, the polymer and
their relative proportions. Adjustments in the material properties will lead to fulfillment of the
requirements for a particular device. [link] shows a comparison among piezoelectric ceramics, polymers
and composites parameters where Z is the impedance, €'33 is the dielectrical constant, and p is the density.

Material parameter Piezoceramics Piezopolymers Piezocomposites
k, (%) 45-55 20 - 30 60 - 75

Z (10° Rayls) 20 - 30 15-4 4-20

€'33/€9 200 - 5000 ~10 50 - 2500

tan y (%) <1 15-5 <1

Qn 10 - 1000 5-10 2-50

p (10° kg/m?) 5.5-8 1-2 2-5

Parameter ranges for piezoelectric ceramics, polymers and composites.

Piezoelectric coatings.

Many potential applications exist which require film thickness of 1 to 30 ym. Some examples of these
macroscopic devices include ultrasonic high frequency transducers, fiber optic modulators and for self
controlled vibrational damping systems.

ZnO and PZT have been used for piezoelectric fiber optic phase modulators fabrication. The piezoelectric
materials have been sputter deposited using dc magnetron source and multimagnetron sputtering systems.
Coatings of 6 ym thick of ZnO and 0.5 jm of PZT are possible to achieve using these systems. However,
thickness variation of approximately 15% occurs between the center and the end of ZnO coatings, results
on affected modulation performance. Although PZT coatings achieved by sputtering posses uniformity and
do not exhibit cracking, the PZT is only partially crystallized and it is actually a composite structure
consisting of crystalline and amorphous material, diminishing the piezoelectric properties.

Sol-gel technique for thick PZT films have been developed. It is now possible to fabricate PZT sol-gel
films of up to 60 pm. The electrical and piezoelectrical properties of the thick films reported are
comparable with ceramic PZT.

Piezoelectric polymer coatings for high-frequency fiber-optic modulators have been also investigated.
Commercial vinylidene fluoride and tetrafluoroethylene copolymer has been used. The advantage of using
polymer coatings is that the polymer jacket (coating) can be easily obtained by melt extrusion on a single-

mode fiber. Thus, uniformity is easily achieved and surface roughness is not present. Furthermore, if
annealing of the polymer is made prior poling, a high degree of crystallinity is enhanced, leading to better
piezoelectric properties.

Bibliography

e R.N. Kleiman, Mat. Res. Soc. Symp. Proc., 1996, 406, 221.

e T. Yamamoto, Jpn. J. Appl. Phys., 1996, 35, 5104.

e Y. Yamashita, Y. Hosono, and N. Ichinose, Jpn. J. Appl. Phys., 1997, 36, 1141.

e I. Akimov and G. K. Savchuk, Inorg. Mater., 1997, 33, 638.

e L. Del Olmo and M. L. Calzada, J. Non-Cryst. Solids, 1990, 121, 424.

e T. Nishi, K. Igarashi, T. Shimizu, K. Koumoto, and H. Yanagida, J. Mater. Sci. Lett., 1989, 8, 805.
e K.R.M. Rao, A. V. P. Rao, and S. Komarneni, Mater. Lett., 1996, 28, 463.

e K. Shimamura, H. Takeda, T. Kohno, and T. Fukuda, J. Cryst. Growth, 1996, 163, 388.
e H. Takeda, K. Shimamura, T. Kohno, and T. Fukuda, J. Cryst. Growth, 1996, 169, 503.
e Lee, T. Itoh and T. Suga, Thin Solid Films, 1997, 299, 88.

e L. J. Mathias, D. A. Parrish, and S. Steadman, Polymer, 1994, 35, 659.

e G.R. Fox, N. Setter, and H.G. Limberger, J. Mater. Res., 1996, 11, 2051.

e L. Gin and B. C. Baxter, Polymer Preprints, 1996, 38, 211.

Formation of Silicon and Gallium Arsenide Wafers

Integrated circuits (ICs) and discrete solid state devices are manufactured on semiconductor
wafers. The following focuses on the general principles and methods with regard to wafer
formation.

Introduction

Integrated circuits (ICs) and discrete solid state devices are manufactured on semiconductor
wafers. Silicon based devices are made on silicon wafers, while III-V (13-15) semiconductor
devices are generally fabricated on GaAs wafers, however, for certain optoelectronic
applications InP wafers are also used. The electrical and chemical properties of the wafer
surface must be well controlled and therefore the preparation of starting wafers is a crucial
portion of IC and device manufacturing. In order to obtain high fabrication yields and good
device performance, it is very important that the starting wafers be of reproducibly high
quality. For example, the front surface must be smooth and flat on both a macro- and
microscale, because high-resolution patterns (lithography) are optically formed on the wafer.
In principle, cutting a crystal into thin slices and polishing one side until all saw marks are
removed and the surface appears smooth and glossy could produce a suitable wafer.
However, due in part to the brittleness of Si and GaAs crystals, as well as the increasing
requirements of wafer cleanliness and surface defect reduction with ever decreasing device
geometries, a very complex series of processing steps are required to produce analytically
clean, flat and damage-free wafer surfaces.

The following focuses on the general principles and methods with regard to wafer formation.
Detailed formulas, recipes, and specific process parameters are not given as they vary
considerably among different wafer producers. However, in general, techniques for
fabrication of Si wafers have generally become standardized within the semiconductor
industry. In contrast, GaAs wafer technology is less standardized, possibly due to either (a)
the similarity to silicon practices or (b) the lower production volume of GaAs wafers. There
are two general classes of processes in the methodology of making wafers: mechanical and
chemical. As both Si and GaAs are brittle materials, the mechanical processes for their wafer
fabrication are similar. However, the different chemistry of Si and GaAs require that the
chemical processes be dealt with separately.

Wafer formation procedures

Each of the processing steps in the conversion of a semiconductor ingot (formed by
Czochralski or Bridgeman growth) into a polished wafer ready for device fabrication, results
in the removal of material from the original ingot; between !/3 and '/, of the original ingot is
sacrificed during processing. Methods for the removal of material from a crystal ingot are
classified depending on the size of the particles being removed during the process. If the
removed particles are much larger than atomic or molecular dimensions the process is
described as being macro-scale. Conversely, if the material is removed atom-by-atom or
molecule-by-molecule then the process is termed micro-scale. A further distinction between
various types of processes is whether the removal occurs as a result of mechanical or

chemical processes. The formation of a finished wafer from a semiconductor ingot normally
requires six machining (mechanical) operations, two chemical operations, and at least one
polishing (chemical-mechanical) operation. Additionally, multiple inspection and evaluation
steps are included in the overall process. A summary of the individual steps, and their
functions, involved in wafer production is shown in [link].

Process Type Function
: : removal of conical shaped ends and impure
cropping mechanical ;
portions
grinding mechanical obtain precise diameter
orientation P identification of crystal orientation and
; mechanical

flatting dopant type
etching chemical removal of surface damage
wafering mechanical formation of individual wafers by cutting
heat treatment thermal annihilation of undesirable electronic donors
edge : : :

: mechanical provide radius on the edge of the wafer
contouring
lapping mechanical provides requisite flatness of the wafer
etching chemical removal of surface damage
polishing pieeMene provides a smooth (specular) surface

chemical

cleaning ee | removal of organics, heavy metals, and

particulates

Summary of the process steps involved in semiconductor wafer production.

Crystal shaping

Although an as-grown crystal ingot is of high purity (99.9999%) and crystallinity, it does not
have the sufficiently precise shape required for ready wafer formation. Thus, prior to slicing
an ingot into individual wafers, several steps are needed. These operations required to prepare
the crystal for slicing are referred to as crystal shaping, and are shown in [link].

(b) (© @ (b)

Schematic representation of crystal shaping operations:
(a) remove crown and taper, (b) grind to required
diameter, (c) grind flat, and (d) slice sample for
measurements. Shaded area represents material
removed.

Cropping

The as-grown ingots have conical shaped seed (top) and tang (bottom) ends that are removed
using a circular diamond saw for ease of further manipulation of the ingot ((link]a). The
cuttings are sufficiently pure that they are cleaned and the recycled in the crystal growth
operation. Portions of the ingot that fail to meet specifications of resistivity are also removed.
In the case of silicon ingots these sections may be sold as metallurgical-grade silicon (MGS).
Conversely, portions of the crystal that meet desired resistivity specifications may be
preferentially selected. A sample slice is also cut to enable oxygen and carbon content to be
determined; usually this is accomplished by Fourier transform infrared spectroscopic
measurements (FT-IR). Finally, cropping is used to cut crystals to a suitable length to fit the
saw Capacity.

Grinding

The primary purpose of crystal grinding is to obtain wafers of precise diameter because the
automatic diameter control systems on crystal growth equipment are not capable of meeting
the tight wafer diameter specifications. In addition, crystals are seldom grown perfectly round
in cross section. Thus, ingots are usually grown with a 1 - 2 mm allowance and reduced to the
proper diameter by grinding [link]b.

Crystal grinding is a straightforward process using an abrasive grinding wheel, however, it
must be well controlled in order to avoid problems in subsequent operations. Exit chipping in
wafering and lattice slip in thermal processing are problems often resulting from improper
crystal grinding. Two methods are used for crystal grinding: (a) grinding on center and (b)
centerless grinding.

[link] shows a schematic of the general set-up for grinding a crystal ingot on center. The
crystal is supported at each end in a lathe-like machine. The rotating cutting tool, employing
a water-based coolant, makes multiple passes down the rotating ingot until the requisite
diameter is obtained. The center grinder can also be used for grinding the identification flats
as well as providing a uniform ingot diameter. However, grinding the crystal on centers
requires that the operator locate the crystal axis in order to obtain the best yield.

relative
movement

diamond cup
wheel

Schematic representation of
grinding on center.

Centerless grinding eliminates the problems associated with locating the crystal center. The
centerless method is superior for long crystals; however, a centerless grinder is much larger
than a center grinder of the same diameter capacity. In centerless grinding the ingot is
supported between two wheels, a grinding wheel and a drive wheel. A schematic of the
centerless grinder is shown in [link]. The axis of the drive wheel is canted with respect to that
of the crystal ingot and the grinding wheel pushing the crystal ingot past the stationary (but
rotating) grinding wheel, see [link]b.

grinding ; drivewheel
wheel drive wheel grinding

movement
of ingot

Schematic representation of centerless grinding viewed
(a) along and (b) perpendicular to the crystal axis.

Orientation/identification flats

Following grinding of the ingot to the desired diameter, one or two flats are ground along the
length of the ingot. The identification flats (one or two) are ground lengthwise along the
crystal according to the orientation and the dopant type. After grinding the crystal on centers
the crystal is rotated to the proper orientation, then the wheel is positioned with its axis of
rotation perpendicular to the crystal axis and moved along the crystal from end to end until
the appropriate flat size is obtained. An optical or X-ray orientation fixture may be used in
conjunction with the crystal mounting to facilitate the proper orientation of the crystal on the
grinder.

The largest flat is called the primary flat ({link]c) and is parallel to one of the crystal planes,
as determined by X-ray diffraction. The primary flat is used for automated positioning of the
wafer during subsequent processing steps, e.g., lithographic patterning and dicing. Other
smaller flats are called "secondary flats" and are used to identify the crystal orientation
(<111> versus <100>) and the material (n-type versus p-type). Secondary flats provide a
quick and easy manner by which unknown wafers can be sorted. The flats shown
schematically in [link] are located according to a Semiconductor Equipment and Materials
Institute (SEMI®) standard and are ground to specific widths, depending upon crystals
diameter. Notches are also used in place of the secondary flat; however, the relative
orientations of the notch and primary flat with regard to crystal orientation and dopant are
maintained.

secondary
flat
<>

secondary | (10° primary ~~ primary

flat flat flat

(100) n-type (100) p-type
secondary
S flat
A] primary primary
flat flat
(111) n-type (111) p-type

SEMI locations for orientation/identification flats.

Etching

The cropping and grinding processes are performed with relatively coarse abrasive and
consequently a great deal of subsurface damage results. Pits, chips, and cracks all contribute
to stress in the cut wafer and provide nuclei for crack propagation at the edges of the finished
wafer. If regions of stress are removed then cracks will no longer propagate, reducing exit
chipping and wafer breakage during subsequent fabrication steps.

The general method for removing surface damage is to etch the crystal in a hot solution. The
most common etchants for Si are based on the HNO3-HF system, in which etchant modifiers
such as acetic acid also commonly used. In the case of GaAs HCI-HNO3 is the appropriate
system. These etchants selectively attack the crystal at the damaged regions. After etching,
the crystal is transferred to the slicing preparation area.

Wafering

The purpose of wafering is to saw the crystal into thin slices with precise geometric
dimensions. By far, the most common method of wafering semiconductor crystals is the use
of an annular, or inner diameter (ID), diamond saw blade. A schematic diagram of ID slicing
technology is shown in [link].

wafer being
sliced

diamond

saw blade cutting edge

blade translation
—$—

Schematic diagrams of ID slicing
process.

The crystal, when it arrives at the sawing area, has been ground to diameter, flatted, and
etched. In order to slice it, the crystal must be firmly mounted in such a way that it can be
completely converted to wafers with minimum waste. The crystal is attached with wax or
epoxy to a mounting block, which is usually cylindrical in shape and of the same diameter as
the ingot. Also, a mounting beam (or strip) is attached along the length of the crystal at the
breakout point of the saw blade. This reduces exit chipping (breakage that occurs as the blade
exits the crystal at the end of a cut) and also provides support for the sawn wafer until it is
retrieved. Graphite or phenolic resins are common materials for the mounting block and
beams, although some success has been obtained in mounting ingots using hydraulic
pressure. The saw blade is a thin sheet of stainless steel (325 um), with diamond bonded to its
inner edge. This blade is mounted on a drum that rotates at ca. 2000 rpm. Saw blades 58 cm
(23 inches) in diameter with a 20 cm (8 inches) opening are common, however, as wafer
sizes increase larger blades are employed: 30 cm (12 inches) wafers are now common for Si.
The blade moves relative to the stationary crystal at a speed of 0.05 cm/s, and the cutting
process is water-cooled. Thus, considering that wafers are sliced sequentially (one at a time),
the overall process is very slow. A further problem is that the kerf loss (loss due to the width
of the blade) results in approximately 1/3 of the material being lost as saw dust. Finally, the
depth of the drum onto which the blade is attached limits the length of the ingot section that
is accessible. In order to overcome this problem, another style of ID blade saw was developed
in which the blade is mounted on an air bearing and is rotated by a belt drive. This allows the
entire length of the crystal ingot to be sliced.

Both silicon and GaAs crystals are grown with either the crystallographic <100> or <111>
direction parallel to the cylindrical axis of the crystal. Wafers may be cut either exactly
perpendicular to the crystallographic axis or deliberately off-axis by several degrees. In order
to obtain the proper wafer orientation, the crystal must be properly oriented on the saw. All
production slicing machines have adjustments for orientation of the crystal; however, it is

usually necessary to check the orientation of the first slice in order to assure that all
subsequent slices will be properly oriented.

Obvious variables introduced during the wafering process include: cutting rate, wheel speed,
and coolant flow rate. However, the condition of the machines, such as alignment and
vibration, is the most important variable followed by the condition of the blade. A deviated
blade rim may cause taper, bow, or warp. [link] summarizes the types of deformations that
can occur during wafering, their physical appearance and their characteristics.

Type of bow and Surface Lattice
warp appearance curvature

Comments

— flat flat ideal

SES curved flat

SSS curved curved

—————4 flat curved

Soe curved flat slips

Deformed wafers and their characteristics.

Heat treatment

As-produced Czochralski grown crystals often have a level of oxygen impurity that may
exceed the concentration of dopant in the semiconductor material (i.e., Si or GaAs). This
oxygen impurity has a deleterious effect on the semiconductor properties, especially upon
subsequent thermal processing, e.g., thermal oxide growth or epitaxial film growth by metal
organic chemical vapor deposition (MOCVD). For example, when silicon crystals are heated

to about 450 °C the oxygen undergoes a transformation that causes it to behave as an electron
donor, much like an n-type dopant. These oxygen donors, or "thermal donors", mask the true
resistivity of the semiconductor because they either add additional carrier electrons to a n-
type crystal or compensate for the positive holes in a p-type crystal. Fortunately, these
thermal donors can be "annihilated" by heat treating the materials briefly in the range of 500 -
800 °C and then cooling quickly through the 450 °C region before donors can reform. In
principle thermal donor annihilation can be performed on wafers at any time during their
fabrication; however, it is usually best to perform the heat treatment immediately after
wafering since sub-standard wafers may be rejected before additional processing steps are
undertaken and thus limiting additional cost. Donor annihilation is a bulk effect, and
therefore the thermal treatment can be performed in air, since any surface oxide that may
form will be removed in subsequent lapping and polishing steps.

Lapping or grinding

The as-cut wafers vary sufficiently in thickness to require an additional operation, the slicing
operation does not consistently produce the required flatness and parallelism required for
many wafer specifications, see [link]. Since conventional polishing does not correct
variations in flatness or thickness, a mechanical two-sided lapping operation is performed.
Lapping is capable of achieving very precise thickness uniformity, flatness and parallelism.
Lapping also prepares the surface for polishing by removing the sub-surface sawing damage,
replacing it with a more uniform and smaller lapping damage.

The process used for lapping semiconductor wafers evolved from the optical lens
manufacturing industry using principles developed over several hundred years. However, as
the lens has a curved surface and the wafers are flat, the equipment for lapping wafers is
mechanically simpler than lens processing machines. The simplest double-side lapping
machine consists of two very flat counter-rotating plates, carriers to hold and move the
wafers between the plates, and a device to feed abrasive slurry steadily between the plates.
The abrasive is typically a 9 pm Al,O3 grit. Commercial abrasives are suspended in water or
glycerin with proprietary additives to assist in suspension and dispersion of the particles, to
improve the flow properties of the slurry, and to prevent corrosion of the lapping machine.
Hydraulics or an air cylinder applies lapping pressure with low starting pressure for 2 to 5
minutes, which is then increased through most of the process. The completion of lapping may
be determined by elapsed time or by an external thickness sensing device. The finished
process gives a wafer with a surface uniform to within 2 pm. Approximately 20 ym per side
is removed during the lapping process.

Although lapping would appear to be simple in concept, the successful implementation of a
production lapping operation requires the development of a technique and experience to
achieve acceptable quality with good yields. Small adjustments to the rotation rates of the
plates and carriers will cause the plates to wear concave, convex or flat.

As lapping is a messy process, various efforts have been made to avoid it or to substitute an
alternative process. The most likely approach at present is grinding, in which the wafer is
held on a vacuum chuck and a series of progressively finer diamond wheels is moved over

the wafer while it is rotated on a turn table. Grinding gives a clearer surface than lapping,
however, only one side may be ground at a time and the resulting flatness is not as good as
that obtained by lapping.

Edge contouring

The rounding of the edge of the wafer to a specific contour is a fairly recent development in
the technology of wafer preparation. It was known by the early seventies that a significant
number of device yield problems could be traced to the physical condition of the wafer edge.
An acute edge affects the strength of the wafer due to: stress concentration, and a lowering of
its resistance to thermal stress, as well as being the source of particle chip, breakage, and
lattice damage. In addition, the particles originating from the chipped edges can, if present on
the wafer surface, add to the defect density (Do) of the IC process reducing fabrication yield.
Further problems associated with a square edge include the build-up of photoresist at the
wafer edge. The solution to these process problems is to provide a contoured edge with a
defined radius (r).

Chemical etching of wafers results in a degree of edge rounding, but it is difficult to control.
Thus, mechanical edge contouring has been developed and the result has been a dramatic
improvement in yields in downstream wafer processing. Losses due to wafer breakage are
also reduced. The edge contouring process is usually performed in cassette-fed high speed
equipment, in which each wafer is rotated rapidly against a shaped cutting tool ({link]).

contoured :
edge cutting tool

a f—— diamonds

Schematic illustration of edge contouring.

Etching

The mechanical processes described above to shape the wafer leave the surface and edges
damaged and contaminated. The depth of the work damage depends on the specific process,
however, 10 um is typical. Such damage is readily removed by chemical etching. Etching is
used at multiple points during the fabrication of a semiconductor device. The discussion
below is limited to etches suitable for wafer fabrication, i.e., non-selective etching of the
entire wafer surface.

Wet chemical etching

The wet chemical etching of any material can be considered to involve three steps: (a)
transportation of the reactants to the surface, (b) reaction at the surface, and (c) movement of
the reaction products into the etchant solution ({link]). Each of these may be the rate limiting
step and thus control the etch rate and uniformity. This effect is summarized in [link].

of

:* Etchant solution’. ‘ reagents .*.

. .
. sotes eset ev et etey

esos er el eter ererere
oretete

cS Semiconductor surface

CLO E EIT

-° Diffusion <*:

* + Diffusion !
- lof reaction -
i+2+ products

Schematic representation of the three steps
involved in wet chemical etching: (i) diffusion of
the chemical etch reagents through the boundary

layer, (ii) chemical reaction at the surface, and (iii)
diffusion of the reaction products into the etch
solution through the boundary layer.

peas Etchin

Rate limiting step pais
rate

Diffusion of reagent to the
surface
Reaction at semiconductor
surface
Diffusion of reaction slow polishing(isotropic)

products from the surface

Results

slow etching(anisotropic)

fast polishing(isotropic)

Comments

enhanced
surface
roughness

ideal

reaction
product

remains on
surface

Effects of rate limiting step in semiconductor etching.

An etchant that is limited by the rate of reaction at the surface will tend to enhance any
surface features and promote surface roughness due to preferential etching at defects
(anisotropic). In contrast, if the etch rate is limited by the diffusion of the etchant reagent
through a stagnant (dead) boundary layer near the surface, then the etch will result in uniform
polishing and the surface will become smooth (isotropic). If removal of the reaction products
is rate limiting then the etch rate will be slow because the etch equilibrium will be shifted
towards the reactants. In the case of an individual etchant reaction, the rate determining step
may be changed by rapid stirring to aid removal of reaction products, or by increasing the
temperature of the etch solution, see [link]. The exact etching conditions are chosen
depending on the application. For example, dilute high temperature etches are often
employed where the etch damage must be minimized, while cooled etches can be used where
precise etch control is required.

100

Etch Rate
(mm.min“!)

10

10 20 30 40 50 60
Temperature (°C)

Typical etch rate versus temperature plot for
a mixture of HF (20%), nitric acid (45%),
and acetic acid (35%).

Traditionally mixtures of hydrofluoric acid (HF), nitric acid (HNO3) and acetic acid
(MeCO>H) have been used for silicon, but alkaline etches using potassium hydroxide (KOH)
or sodium hydroxide (NaOH) solutions are increasingly common. Similarly, gallium arsenide
etches may be either acidic or basic, however, in both cases the etches are oxidative due to
the use of hydrogen peroxide. A wide range of chemical reagents are commercially available
in "transistor grade" purity and these are employed to minimize contamination of the

semiconductor. Deionized water is commonly used as a diluent for each of these reagents and
the concentration of commonly used aqueous reagents is given in [link].

Reagent Weight % Reagent Weight %
HCl 37 HF 49

H SO, 98 H3PO4 85

HNO; 79 HClO, 70
MeCO,H 99 H 05 30
NH,OH 29

Weight percent concentration of commonly used concentrated aqueous reagents.

The equipment used for a typical etchant process includes an acid (or alkaline) resistant tank,
which contains the etchant solution and one or more positions for rinsing the wafers with
deionized water. The process is batch in nature involving tens of wafers and the best
equipment provides a means of rotating the wafers during the etch step to maintain
uniformity. In order to assure the removal of all surface damage, substantial over-etching is
performed. Thus, the removal of 20 1m from each side of the wafer is typical. Etch times are
usually several minutes per batch.

Etching silicon

The most commonly used etchants for silicon are mixtures of hydrofluoric acid (HF) and
nitric acid (HNO3) in water or acetic acid (MeCO>H). The etching involves a reduction-
oxidation (redox) reaction, followed by dissolution of the reaction products. In the HF-HNO3
system the HNO3 oxidizes the silicon and the HF removes the reaction products from the
surface. The overall reaction is:

Equation:

Si + HNO, + 6 HF > H,SiF, + HNO, + H,O

The oxidation reaction involves the oxidation of Si to Si**, and it is auto-catalytic in that the
reaction product promotes the reaction itself. The initial step involves trace impurities of

HNO, in the HNO; solution, [link], which react to liberate nitrogen dioxide (NO>), [link].
Equation:

HNO, + HNO, > NO, +H;0
Equation:

N,O, > 2.NO,

The nitrogen dioxide oxidizes the silicon surface in the presence of water, resulting in the
formation of Si(OH)» and the reformation of HNOb, [link]. The Si(OH) decomposes to give
SiO», [link]. Since the reaction between HNO» and HNOs, [link], is rate limiting, an
induction period is observed. However, this is overcome by the addition of NO. ions in the
form of [NH,][NO>].

Equation:

Si’ +2 NO, +2 H,O > Si(OH), + 2 HNO,
Equation:

Si(OH), > SiO, + H,

The final step of the etch process is the dissolution of the SiO» by HF, [link]. Stirring serves
to remove the soluble products from the reaction surface. The role of the HF is to act as a
complexing reagent, and thus the reaction shown in [link] is known as a complexing reaction.
The formation of water as a reaction product requires that acetic acid be used as a diluent
(solvent) to ensure better control.

Equation:

SiO, +6 HF > H,SiF, + H,O

The etching reaction is highly dependent on the relative ratios of the etchant reagents. Thus,
if an HF-rich solution is used, the reaction is limited by the oxidation step, [link], and the
etching is anisotropic, since the oxidation reaction is sensitive to doping, crystal orientation,
and defects. In contrast, the use of a HNO3-rich solution produces isotropic etching since the
dissolution process is rate limiting ({link]). The reaction of HNO3-rich solutions has been
found to be diffusion-controlled over the temperature range 20 - 50 °C ([link]), and is
therefore commonly employed for removing work damage produced during wafer
fabrication. The boundary layer thickness ({link]) and therefore the dimensional control over
the wafer is controlled by the rotation rate of the wafers. A common etch formulation is a
4:1:3 mixture of HNO3 (79%), HF (49%), and MeCO>H (99%). There are some etchant

formulations that are based on alternative (or additional) oxidizing agents, such as: Bro, Lb,
and KMnO,.

Alkaline etching (KOH/H20 or NaOH/H)0) is by nature anisotropic and the etch rate
depends on the number of dangling bonds which in turn are dependent on the surface
orientation. Since etching is reaction rate limited no rotation of the wafers is necessary and
excellent uniformity over large wafers is obtained. Alkaline etchants are used with large
wafers where dimensional uniformity is not maintained during lapping. A typical formulation
uses KOH in a 45% weight solution in HzO at 90 °C.

Etching gallium arsenide

Although a wide range of etches have been investigated for GaAs, few are truly isotropic.
This is because the surface activity of the (111) Ga and (111) As faces are very different. The
As rich face is considerably more reactive than the Ga rich face, thus under identical
conditions it will etch faster. As a result most etches give a polished surface on the As face,
but the Ga face tends to appear cloudy or frosted due to the highlighting of surface features
and crystallographic defects.

As with silicon the etch systems involve oxidation and complexation. However, in the case of
GaAs the gallium is already fully oxidized (formally Ga**), thus, it is the arsenic (formally
the arsenide ion, As* that is oxidized by a suitable oxidizing agent (e.g., HO>) to the soluble
oxide, As»O3, [link]. The gallium ions form the oxide Ga)O3 via the hydroxide, [link]. Both
oxides are soluble in acid solutions, resulting in their removal from the surface.

Equation:

2 As* +6 H,O, + H* > As,O, +5 OH +4H,0
Equation:

2 Ga** + 6 OH > 2 Ga(OH); +3 H,O

The peroxide based oxidative etches for GaAs are divided into acidic and basic etches. The
composition and application of some of these systems are summarized in [link]. The most
widely used of these is HySO,4/H»O>/H>O and is referred to as Caro's acid. The high viscosity
of H2SO, results in diffusion-limited etching with high acid concentrations. Etches with low
acid concentrations tend to be anisotropic. Phosphoric acid (H3PO,) or citric acid ((link])
may be exchanged for sulfuric acid (H»SO,). Replacement of the acid component with bases
such as NH,OH or NaOH can result in near to truly isotropic etchants, although certain
combinations can result in strong anisotropy.

Formulation volume etch rate etch rate etch rate
ratio (um/min) (m/min) (m/min)
ae 5 3:1:50 0.8 0.8 0.8
ae Lid 0.6 0.6 0.6
ne . 1:700 0.3 0.3 0.3
ee " 1:0.76 0.2 0.2 0.2

The composition and application of selected etch systems for GaAs.

Ox OH
| | I
HO” CH; | > OH
OH

Structure of citric
acid.

(100) (110) (111)As

(111)Ga
etch rate
(m/min)

0.8

3.0

0.4

0.4

0.3

0.2

One of the earliest etching systems for GaAs is based on the use of a dilute (ca. 0.05 vol.%)
solution of bromine (Br>) in ethanol. The Bry acts as the oxidant, resulting in the formation of
soluble bromides. The etch rate of this system is different for different crystallographic
planes, i.e., the etch rates for the (111) As, (100), and (111) Ga faces are in the ratio 6:5:1,
although more uniform etch rates are observed with high Br concentrations (ca. 10 vol.%).
These higher concentration solutions are used for the removal of damage due to cutting with

the saw.

Polishing

The purpose of polishing is to produce a smooth, specular surface on which device features
can be defined by lithography. In order to allow for very large scale integration (VLSI) or
ultra large scale integration (ULSI) fabrication the wafer must have a surface with a high
degree of flatness. Variations less than 5 to 10 ym across the wafer diameter are typical
flatness specifications. In addition, given the preceding steps, wafer polishing must not leave
residual contamination or surface damage. The techniques of wafer polishing are derived
from the glass lens industry, with some important modifications that have been developed to
meet the special requirements of the microelectronics industry.

Differences between polishing and lapping

If the surface of a wafer that has undergone lapping (or grinding) is examined with an
electron microscope, cracks, ridges and valleys are observed. The top "relief layer" consists
of peaks and valleys. Below this layer is a damaged layer characterized by microcracks,
dislocations, slip and stress. [link] shows a schematic representation of the abraded surface.
Both of these layers must be removed completely prior to further fabrication. Decreasing the
particle size of the abrasive during lapping only decreases the scale of the damage, but does
not eliminate it entirely. In fact this surface damage is a characteristic of the brittle fracture of
single crystal Si and GaAs, and occurs because during lapping the abrasive grains are moved
across the surface under a pressure beyond that of the fracture strength of the wafer materials
(Si or GaAs). In contrast to the mechanical abrasion employed in lapping, polishing is a
mechano-chemical process during which brittle fracture does not occur. A polished wafer
does not display any evidence of a relief surface such as that produced by lapping, even at
highest resolution electron microscope.

relief layer

damaged layer

Jo material

Schematic representation of a cross sectional view of an
abraded wafer surface prior to polishing.

Process of Polishing

[link] shows a schematic of the polishing process. Polishing may be conducted on single
wafers or as a batch process depending on the equipment employed. Single wafer polishing is
preferred for larger wafers and allows for better surface flatness. In both processes, wafers are
mounted onto a fixture, by either wax or a composite Felx-Mount™, and pressed against the
polishing pad. The polishing pad is usually made from an artificial fabric such as polyester
felt-polyurethane laminate. Polishing is accomplished by a mechano-chemical process in
which aqueous polishing slurry is dripped onto the polishing pad, see [link]. The polishing
slurry performs both a chemical and mechanical process, and consists of fine silica (SiO>)
particles (100 A diameter) and an oxidizing agent. Aqueous sodium hydroxide (NaOH) is
used for Si, while aqueous sodium chlorate (NaOCl) is preferred for GaAs. Suspending
agents are usually added to prevent settling of the silica particles. Under the heat caused by
the friction of the wafer on the polishing pad the wafer surface is oxidized, which is the
chemical step, while in the mechanical step the silica particles in the slurry abrade the
oxidized surface away.

pressure

polishing pad

Schematic representation of the wafer
polishing process.

In order to achieve a reasonable rate of removal of the relief and damaged layers and still
obtain the highest quality surface, the polishing is done in two steps, stock removal and haze
removal. The former is carried out with a higher concentration slurry and may proceed for
about 30 minutes at a removal rate of 1 pm/min. Haze removal is performed with a very
dilute slurry, a softer pad with a reaction time of about 5 to 10 minutes, during which the total
amount of material removed is only about 1 jm. Due to the active chemical reaction between
the wafer and the polishing agent, the wafers must be rinsed in deionized water immediately
after polishing to prevent haze or stains from reforming.

There are many variables that will influence the rate and quality of polishing. High pressure
results in a higher polishing rate, but excessive pressure may cause non-uniform polishing,
excessive heat generation and fast pad wear. The rate of polishing is increased with higher
temperatures but this may also lead to haze formation. High wheel speeds accelerate the

polishing rate but can raise the temperature and also results in problems in maintaining a
uniform flow of slurry across the pad. Dense slurry concentrations increase the polishing rate
but are more costly. The pH of the slurry solution can also affect the polishing rate, for
example the polishing rate of Si gradually increases with increased pH (higher basicity) until
a pH of about 12 where a dramatic decrease is observed. In general, the optimum polishing
process for a given facility depends largely upon the interplay of product specification,
yields, cost, and quality considerations and must be developed uniquely. The wafer polishing
process does not improve the wafer flatness and, at best, polishing will not degrade the wafer
flatness achieved in the lapping operation.

Cleaning

During the processes described above, semiconductor wafers are subjected to physical
handling that leads to significant contamination. Possible sources of physical contamination
include:

a. airborne bacteria,

b. grease and wax from cutting oils and physical handling,

c. abrasive particulates (usually, silica, silicon carbide, alumina, or diamond dust) from
lapping, grinding or sawing operations,

d. plasticizers which are derived from containers and wrapping in which the wafers are
handled and shipped.

Chemical contamination may also occur as a result of improper cleaning after etch steps.
Light-metal (especially sodium and potassium) species may be traced to impurities in etchant
solutions and are chemisorbed on to the surface where they are particularly problematical for
metal oxide semiconductor (MOS) based devices, although higher levels of such impurities
are tolerable for bipolar devices. Heavy metal impurities (e.g., Cu, Au, Fe, and Ag) are
usually caused by electrodeposition from etchant solutions during fabrication. While wafers
are cleaned prior to shipping, contamination accumulated during shipping and storage
necessitates that all wafers be subjected to scrupulous cleaning prior to fabrication.
Furthermore, cleaning is required at each step during the fabrication process. Although wafer
cleaning is a vital part of each fabrication step, it is convenient to discuss cleaning within the
general topic of wafer fabrication.

Cleaning silicon

The first step in cleaning a Si wafer is removal of all physical contaminants. These
contaminates are removed by rinsing the wafer in hot organic solvents such as 1,1,1-
trichloroethane (Cl3CH3) or xylene (CgH,Me>), accompanied by mechanical scrubbing,
ultrasonic agitation, or compressed gas jets. Removal of the majority of light metal
contaminants is accomplished by rinsing in hot deionized water, however, complete removal
requires a further more aggressive cleaning process. The most widely used cleaning method

in the Si semiconductor industry is based on a two step, two solution sequence known as the
“RCA Cleaning Method”.

The first solution consists of HyO-H»O -NH,OH in a volume ratio of 5:1:1 to 7:2:1, which is
used to remove organic contaminants and heavy metals. The oxidation of the remaining
organic contaminants by the hydrogen peroxide (H2O>) produces water soluble products.
Similarly, metal contaminants such as cadmium, cobalt, copper, mercury, nickel, and silver
are solubilized by the NH,OH through the formation of soluble amino complexes, e.g., [link].
Equation:

2 Cu**(s) + 6 NH,OH (aq) > [Cu(NH;)¢ (aq)

The second solution consists of HyO-H 0 -HCI in a 6:1:1 to 8:2:1 volume ratio and removes
the Group [(1), II(2) and III(13) metals. In addition, the second solution prevents re-
deposition of the metal contaminants. Each of the washing steps is carried out for 10 - 20
min. at 75 - 85 °C with rapid agitation. Finally, the wafers are blown dry under a stream of
nitrogen gas.

Cleaning GaAs

In principle GaAs wafers may be cleaned in a similar manner to silicon wafers. The first step
involves successive cleaning with hot organic solvents such as 1,1,1-trichloroethane, acetone,
and methanol, each for 5-10 minutes. GaAs wafers cleaned in this manner may be stored
under methanol for short periods of time.

Most cleaning solutions for GaAs are actually etches. A typical solution is similar to the
second RCA solution and consists of an 80:10:1 ratio of HyO-HO,-HCI. This solution is
generally used at elevated temperatures (70 °C) with short dip times since it has a very fast
etch rate (4.0 m/min).

Measurements, inspections and packaging

Quality control measurements of the semiconductor crystal and subsequent wafer are
performed throughout the process as an essential part of the fabrication of wafers. From
crystal and wafer shaping through the final wafer finishing steps, quality control
measurements are used to ensure that the materials meets customer specifications, and that
problems can be corrected before they create scrap material and thus avoid further processing
of reject material. Quality control measurements can be broadly classified into mechanical,
electrical, structural, and chemical.

Mechanical measurements are concerned with the physical dimensions of the wafer,
including: thickness, flatness, bow, taper and edge contour. Electrical measurements usually
include: resistivity and lateral resistivity gradient, carrier type and lifetime. Measurements

giving information on the perfection of the semiconductor crystal lattice are classified in the
structural category and include: testing for stacking faults, and dislocations. Routine chemical
measurements are limited to the measurement of dissolved oxygen and carbon by Fourier
transform infrared spectroscopy (FT-IR). Finished wafers are individually marked for the
purpose of identification and traceability. Packaging helps protect the finished wafers from
contamination during shipping and storage.

Industry standards defining in detail how quality control measurements are to be made and
determining the acceptable ranges for measured values have been developed by the American
Society of Testing Materials (ASTM) and the Semiconductor Equipment and Materials
Institute (SEMI).

Bibliography

e A.C. Bonora, Silicon Wafer Process Technology: Slicing, Etching, Polishing,
Semiconductor Silicon 1977, Electrochem. Soc., Pennington, NJ (1977).

e L. D. Dyer, in Proceeding of the low-cost solar array wafering workshop 1981, DoE-
JPL-21012-66, Jet Propulsion Lab., Pasadena CA (1982).

e J.C. Dyment and G. A. Rozgonyi, J. Electrochem. Soc., 1971, 118, 1346.

e H. Gerischer and W. Mindt, Electrochem. Acta, 1968, 13, 1329.

P. D. Green, Solid State Electron., 1976, 19, 815.

e C. A. Harper and R. M. Sompson, Electronic Materials & Processing Handbook,

McGraw Hill, New York, 2nd Edition.

S. lida and K. Ito, J. Electrochem. Soc., 1971, 118, 768.

e W. Kern, J. Electrochem. Soc., 1990, 137, 1887.

e Y. Mori and N. Watanabe, J. Electrochem. Soc., 1978, 125, 1510.

D. L. Partin, A. G. Milnes, and L. F. Vassamillet, J. Electrochem. Soc., 1979, 126, 1581.

D. W. Shaw, J. Electrochem. Soc., 1966, 113, 958.

e F, Snimura, Semiconductor Silicon Crystal Technology, Academic Press, New York
(1989).

e D.R. Turner, J. Electrochem. Soc., 1960, 107, 810.

Doping

Starting with a prepared, polished wafer then how do we get an integrated
circuit? We will focus on the CMOS process, described in the last chapter.
Let's assume we have wafer which was doped during growth so that it has a
background concentration of acceptors in it (i.e. it is p-type). Referring back
to CMOS Logic, you can see that the first thing we need to build is a n-tank
or moat. In order to do this, we need some way in which to introduce
additional impurities into the semiconductor. There are several ways to do
this, but current technology relies almost exclusively on a technique called
ion implantation. A diagram of an ion-implanter is shown in the figure in
the previous section. An ion implanter uses a dopant source gas, ionizes it,
and drives the ions into the wafer. The dopant gas is ionized and the
resultant charged ions are accelerated through a magnetic field, where they
are mass-analyzed. The vertical magnetic field causes the beam of ions to
spread out, according to their mass. A thin aperture selects the ions of
interest, and lets them pass, blocking all the others. This makes sure we are
only implanting the ion we want, and in fact, even selects for the proper
isotope! The ionized atoms are then accelerated through several tens to
hundreds of kV, and then deflected by an electric field, much like in an
oscilloscope CRT. In fact, most of the time the ion beam is "rastered" across
the surface of the silicon wafer. The ions strike the silicon wafer and pass
into its interior. A measurement of the current flow in the system and its
integral, is a measure of how much dopant was deposited into the wafer.
This is usually given in terms of the number of dopant stows. to which the

wafer has been exposed.

After the atoms enter the silicon, they interact with the lattice, creating
defects, and slowing down until finally they stop. Typical atomic
distributions, as a function of implant voltage are show in [link] for
implantation into amorphous silicon. When implantation is done on single
crystal material, channeling, the improved mobility of an ion down the
"hallway" of a given lattice direction, can skew the impurity distribution
significantly. Just slight changes of less than a degree can make big
differences in how the impurity atoms are finally distributed in the wafer.
Usually, the operator of the implant machine purposely tilts the wafer a few

degrees off normal to the beam in order to arrive at more reproducible
results.

a
=

102° 25kV 100 kV 300 kV

distance into
wafer

Impurity Concentration c
r=)

1 um

Implant distribution with
acceleration energy

As you might expect, shooting 100 kV ions at a silicon wafer probably does
quite a bit of damage to the crystal structure. Not only that, but just having,
say boron, in your wafer does not mean you are going to have holes. For the
boron to become "electrically active" - that is to act as an acceptor - it has to
reside on a silicon lattice site. Even if the boron atom does, somehow, end
up on an actual lattice site when it stops crashing around in the wafer, the
many defects which have been created will act as deep traps. Thus, the hole
which is formed will probably be caught at a trap site and will not be able to
contribute to electrical conductivity in the wafer anyway. How can we fix
this situation? If we carefully heat up the wafer, we can cause the atoms in
the crystal to shake around, and if we do it right, they all get back where
they belong. Not only that, but the newly added impurities end up on lattice
sites as well! This step is called annealing and it does just what it is
supposed to. Typical temperatures and times for such an anneal are 500 to
1000°C for 10 to 30 minutes.

Something else occurs during the anneal step however. We have just added,
by our implantation step, impurities with a fairly tight distribution as shown
in [link]. There is an obvious gradient in impurity distribution, and if there

is a gradient, than things may start moving around by diffusion, especially
at elevated temperatures.

Applications for Silica Thin Films

Introduction

While the physical properties of silica make it suitable for use in protective
and optical coating applications, the biggest application of insulating SiO»
thin films is undoubtedly in semiconductor devices, in which the insulator
performs a number of specific tasks, including: surface passivation, field
effect transistor (FET) gate layer, isolation layers, planarization and
packaging.

The term insulator generally refers to a material that exhibits low thermal or
electrical conductivity; electrically insulating materials are also called
dielectrics. It is in regard to the high resistance to the flow of an electric
current that SiO» thin films are of the greatest commercial importance. The
dielectric constant (€) is a measure of a dielectric materials ability to store
charge, and is characterized by the electrostatic energy stored per unit
volume across a unit potential gradient. The magnitude of ¢ is an indication
of the degree of polarization or charge displacement within a material. The
dielectric constant for air is 1, and for ionic solids is generally in the range
of 5 - 10. Dielectric constants are defined as the ratio of the material’s
capacitance to that of air, i.e., [link]. The dielectric constant for silicon
dioxide ranges from 3.9 to 4.9, for thermally and plasma CVD grown films,
respectively.

Equation:

& = Comerai/C

material air

An insulating layer is a film or deposited layer of dielectric material
separating or covering conductive layers. Ideally, in these application an
insulating material should have a surface resistivity of greater than 10!%
Q/cm? or a volume resistivity of greater than 10!! Q.cm. However, for
some applications, lower values are acceptable; an electrical insulator is
generally accepted to have a resistivity greater than 10° Q.cm. CVD SiO,
thin films have a resistivity of 10° - 10'° Q.cm, depending on the film
growth method.

As a consequence of its dielectric properties SiO», and related silicas, are
used for isolating conducting layers, to facilitate the diffusion of dopants
from doped oxides, as diffusion and ion implantation masks, capping doped
films to prevent loss of dopant, for gettering impurities, for protection
against moisture and oxidation, and for electronic passivation. Of the many
methods used for the deposition of thin films, chemical vapor deposition
(CVD) is most often used for semiconductor processing. In order to
appreciate the unique problems associated with the CVD of insulating SiO»
thin films it is worth first reviewing some of their applications. Summarized
below are three areas of greatest importance to the fabrication of
contemporary semiconductor devices: isolation and gate insulation,
passivation, and planarization.

Device isolation and gate insulation

A microcircuit may be described as a collection of devices each consisting
of "an assembly of active and passive components, interconnected within a
monolithic block of semiconducting material". Each device is required to be
isolated from adjacent devices in order to allow for maximum efficiency of
the overall circuit. Furthermore within a device, contacts must also be
electrically isolated. While there are a number of methods for isolating
individual devices within a circuit (reverse-biased junctions, mesa isolation,
use of semi-insulating substrates, and oxide isolation), the isolation of the
active components in a single device is almost exclusively accomplished by
the deposition of an insulator.

In [link] is shown a schematic representation of a silicon MOSFET (metal-
oxide-semiconductor field effect transistor). The MOSFET is the basic
component of silicon-CMOS (complimentary metal-oxide-semiconductor)
circuits which, in turn, form the basis for logic circuits, such as those used
in the CPU (central processing unit) of a modern personal computer. It can
be seen that the MOSFET is isolated from adjacent devices by a reverse-
biased junction (p*-channel stop) and a thick oxide layer. The gate, source
and drain contact are electrically isolated from each other by a thin
insulating oxide. A similar scheme is used for the isolation of the collector
from both the base and the emitter in bipolar transistor devices.

source gate drain
contact contact contact

contact metal

YY), Yfyy);

BPSG
tCaxzx—IZ

gate metal ___ . ... thin isolation oxide
thick oxide ~ <n 1 thin oxide, gate

Pt channel stop

source drain

Schematic diagrams of a Si- MOSFET (metal-oxide-semiconductor
field effect transistor).

As a transistor, a MOSFET has many advantages over alternate designs.
The key advantage is low power dissipation resulting from the high
impedance of the device. This is a result of the thin insulation layer between
the channel (region between source and drain) and the gate contact, see
[link]. The presence of an insulating gate is characteristic of a general class
of devices called MISFETs (metal-insulator-semiconductor field effect
transistor). MOSFETs are a subset of MISFETs where the insulator is
specifically an oxide, e.g., in the case of a silicon MISFET device the
insulator is SiO», hence MOSFET. It is the fabrication of MOSFET circuits
that has allowed silicon technology to dominate digital electronics (logic
circuits). However, increases in computing power and speed require a
constant reduction in device size and increased complexity in device
architecture.

Passivation

Passivation is often defined as a process whereby a film is grown on the
surface of a semiconductor to either (a) chemically protect it from the
environment, or (b) provide electronic stabilization of the surface.

From the earliest days of solid state electronics it has been recognized that
the presence or absence of surface states plays a decisive role in the
usefulness of any semiconducting material. On the surface of any solid state
material there are sites in which the coordination environment of the atoms
is incomplete. These sites, commonly termed "dangling bonds", are the
cause of the electronically active states which allow for the recombination
of holes and electrons. This recombination occurs at energies below the
bulk value, and interferes with the inherent properties of the semiconductor.
In order to optimize the properties of a semiconductor device it is desirable
to covalently satisfy all these surface bonds, thereby shifting the surface
states out of the band gap and into the valence or conduction bands.
Electronic passivation may therefore be described as a process which
reduces the density of available electronic states present at the surface of a
semiconductor, thereby limiting hole and electron recombination
possibilities. In the case of silicon both the native oxide and other oxides
admirably fulfill these requirements.

Chemical passivation requires a material that inhibits the diffusion of
oxygen, water, or other species to the surface of the underlying
semiconductor. In addition, the material is ideally hard and resistant to
chemical attack. A perfect passivation material would satisfy both
electronic and chemical passivation requirements.

Planarization

For the vast majority of electronic devices, the starting point is a substrate
consisting of a flat single crystal wafer of semiconducting material. During
processing, which includes the growth of both insulating and conducting
films, the surface becomes increasingly non-planar. For example, a gate
oxide in a typical MOSFET (see [link]) may be typically 100 - 250 A thick,
while the isolation or field oxide may be 10,000 A. In order for the
successful subsequent deposition of conducting layers (metallization) to
occur without breaking metal lines (often due to the difficulty in
maintaining step coverage), the surface must be flat and smooth. This
process is called planarization, and can be carried out by a technique known
as sacrificial etchback. The steps for this process are outlined in [link]. An
abrupt step ([link]a) is coated with a conformal layer of a low melting

dielectric, e.g., borophosphorosilicate glass, BPSG ({link |b), and
subsequently a sacrificial organic resin ([link]c). The sample is then plasma
etched such that the resin and dielectric are removed at the same rate. Since
the plasma etch follows the contour of the organic resin, a smooth surface is
left behind ({link]d). The planarization process thus reduces step height
differentials significantly. In addition regions or valleys between individual
metallization elements (vias) can be completely filled allowing for a route
to producing uniformly flat surfaces, e.g., the BPSG film shown in [link].

metallization
Si-substrate
(a)

CVD
silicate glass

(b)

organic
resist

(c)

(d)

Schematic representation of the planarization process.

A metallization feature (a) is CVD covered with
silicate glass (b), and subsequently coated with an
organic resin (c). After etching the resist a smooth

silicate surface is produced (d).

The processes of planarization is vital for the development of multilevel
structures in VLSI circuits. To minimize interconnection resistance and
conserve chip area, multilevel metallization schemes are being developed in
which the interconnects run in 3-dimensions.

Bibliography

e J. L. Vossen and W. Kern, Phys. Today, 1980, 33, 26.

e S.K. Ghandhi, VLSI Fabrication Principles, Silicon and Gallium
Arsenide, Wiley, Chichester, 2nd Ed. (1994).

e S.M. Sze, Physics of Semiconductor Devices, 2nd Edition, John Wiley

& Sons, New York (1981).

W. E. Beadle, J. C. C. Tsai, R. D. Plummer, Quick Reference Manual

for Silicon Integrated Cuircuit Technology, Wiley, Chichester (1985).

e A.C. Adams and C. D. Capio, J. Electrochem. Soc., 1981, 128, 2630.

Oxidation of Silicon

Note:This module was developed as part of the Rice University course
CHEM-496: Chemistry of Electronic Materials. This module was prepared
with the assistance of Andrea Keys.

Introduction

In the fabrication of integrated circuits (ICs), the oxidation of silicon is
essential, and the production of superior ICs requires an understanding of
the oxidation process and the ability to form oxides of high quality. Silicon
dioxide has several uses:

1. Serves as a mask against implant or diffusion of dopant into silicon.
2. Provides surface passivation.

3. Isolates one device from another (dielectric isolation).

4, Acts as a component in MOS structures.

5. Provides electrical isolation of multi-level metallization systems.

Methods for forming oxide layers on silicon have been developed,
including thermal oxidation, wet anodization, chemical vapor deposition
(CVD), and plasma anodization or oxidation. Generally, CVD is used when
putting the oxide layer on top of a metal surface, and thermal oxidation is
used when a low-charge density level is required for the interface between
the oxide and the silicon surface.

Oxidation of silicon

Silicon's surface has a high affinity for oxygen and thus an oxide layer
rapidly forms upon exposure to the atmosphere. The chemical reactions
which describe this formation are:

Equation:

Equation:

In the first reaction a dry process is utilized involving oxygen gas as the
oxygen source and the second reaction describes a wet process which uses
steam. The dry process provides a "good" silicon dioxide but is slow and
mostly used at the beginning of processing. The wet procedure is
problematic in that the purity of the water used cannot be guaranteed to a
suitable degree. This problem can be easily solved using a pyrogenic
technique which combines hydrogen and oxygen gases to form water vapor
of very high purity. Maintaining reagents of high quality is essential to the
manufacturing of integrated circuits, and is a concern which plagues each
step of this process.

The formation of the oxide layer involves shared valence electrons between
silicon and oxygen, which allows the silicon surface to rid itself of
"dangling" bonds, such as lone pairs and vacant orbitals, [link]. These
vacancies create mid-gap states between the valence and conduction bands,
which prevents the desired band gap of the semiconductor. The Si-O bond
strength is covalent (strong), and so can be used to achieve the loss of mid-
gap states and passivate the surface of the silicon.

NANANANANZ

PPP WWW
P9VRROP ov, VRPOO?D

Si Si Si Si Si Si Si Si Si Si SiS

Removal of dangling bonds by oxidation of surface.

The oxidation of silicon occurs at the silicon-oxide interface and consists of
four steps:

Diffusive transport of oxygen across the diffusion layer in the vapor phase
adjacent to the silicon oxide-vapor interface.

Incorporation of oxygen at the outer surface into the silicon oxide film.
Diffusive transport across the silicon oxide film to its interface with the
silicon lattice.

Reaction of oxygen with silicon at this inner interface.

As the Si-SiO> interface moves into the silicon its volume expands, and
based upon the densities and molecular weights of Si and SiO», 0.44 A Si is
used to obtain 1.0 A SiOp.

Pre-oxidation cleaning

The first step in oxidizing a surface of silicon is the removal of the native
oxide which forms due to exposure to open air. This may seem redundant to
remove an oxide only to put on another, but this is necessary since
uncertainty exists as to the purity of the oxide which is present. The
contamination of the native oxide by both organic and inorganic materials
(arising from previous processing steps and handling) must be removed to
prevent the degradation of the essential electrical characteristics of the
device. A common procedure uses a HyO-H»O,-NH,OH mixture which
removes the organics present, as well as some group I and II metals.
Removal of heavy metals can be achieved using a H»O-H»O>-HCI mixture,
which complexes with the ions which are formed. After removal of the
native oxide, the desired oxide can be grown. This growth is useful because
it provides: chemical protection, conditions suitable for lithography, and
passivation. The protection prevents unwanted reactions from occurring and
the passivation fills vacancies of bonds on the surface not present within the
interior of the crystal. Thus the oxidation of the surface of silicon fulfills
several functions in one step.

Thermal oxidation

The growth of oxides on a silicon surface can be a particularly tedious
process, since the growth must be uniform and pure. The thickness wanted
usually falls in the range 50 - 500 A, which can take a long time and must
be done on a large scale. This is done by stacking the silicon wafers in a
horizontal quartz tube while the oxygen source flows over the wafers,
which are situated vertically in a slotted paddle (boat), see [link]. This
procedure is performed at 1 atm pressure, and the temperature ranges from
700 to 1200 °C, being held to within +1 °C to ensure uniformity. The choice
of oxidation technique depends on the thickness and oxide properties
required. Oxides that are relatively thin and those that require low charge at
the interface are typically grown in dry oxygen. When thick oxides are
required (> 0.5 mm) are desired, steam is the source of choice. Steam can
be used at wide range of pressures (1 atm to 25 atm), and the higher
pressures allow thick oxide growth to be achieved at moderate temperatures
in reasonable amounts of time.

‘ian tube

"©"
—_

silicon
wafers

Horizontal diffusion tube showing the
oxidation of silicon wafers at 1 atm
pressure.

The thickness of SiO> layers on a Si substrate is readily determined by the
color of the film. [link] provides a guidline for thermal grown oxides.

Film
thickness
(jum)

0.05

0.07

0.10

0.12

0.15

0.17

0.20

0.22
0.25

0.27

0.30

0.31

0.32

0.34

Color

tan
brown

dark violet to
red-violet

royal blue

light blue to
metallic blue

metallic to light

yellow-green

light gold

gold

orange to melon

red-violet

blue to violet
blue

blue

blue to blue-
green

light green

Film
thickness
(ym)

0.63

0.68

0.72

0.77

0.80

0.82

0.85

0.86
0.87

0.89

0.92

0.95

0.97

0.99

Color

violet-red
"bluish"

blue-green to
gree

"yellowish"

orange

salmon

light red-
violet

violet
blue violet

blue

blue-green

yellow-green

yellow

orange

0.35

0.36

0.37

0.39

0.41

0.42

0.44

0.46

0.47

0.48

0.49

0.50

0.52

0.54

0.56

0.57

green to yellow-

green
yellow-green
green-yellow
yellow

light orange
carnation pink
violet-red
red-violet
violet
blue-violet

blue

blue green

green
yellow-green

green-yellow

"yellowish"

1.00

1.02

1.05

1.06

1.07

1.10

1.11

1.12

1.18

1.19

1.21

1.24

1325

1.28

1.32

1.40

carnation
pink

violet red
red-violet
violet
blue-violet
green
yellow-green
green
violet
red-violet
violet-red
carnation
pink to
salmon
orange

"yellowish"

sky blue to
green-blue

orange

0.58 light orange to 1.46 blue-violet
pink

0.60 carnation pink 1.50 blue

Color chart for thermally grown SiO, films observed under daylight
fluorescent lighting.

High pressure oxidation

High pressure oxidation is another method of oxidizing the silicon surface
which controls the rate of oxidation. This is possible because the rate is
proportional to the concentration of the oxide, which in turn is proportional
to the partial pressure of the oxidizing species, according to Henry's law,
[link], where C is the equilibrium concentration of the oxide, H is Henry's
law constant, and pg is the partial pressure of the oxidizing species.
Equation:

C = Hg)

This approach is fast, with a rate of oxidation ranging from 100 to 1000
mm/h, and also occurs at a relatively low temperature. It is a useful process,
preventing dopants from being displaced and also forms a low number of
defects, which is most useful at the end of processing.

Plasma oxidation

Plasma oxidation and anodization of silicon is readily accomplished by the
use of activated oxygen as the oxidizing species. The highly reactive
oxygen is formed within an electrical discharge or plasma. The oxidation is
carried out in a low pressure (0.05 - 0.5 Torr) chamber, and the the plasma
is produced either by a DC electron source or a high-frequency discharge.
In simple plasma oxidation the sample (i.e., the silicon wafer) is held at

ground potential. In contrast, aniodization systems usually have a DC bias
between the sample and an electrode with the sample biased positively with
respect to the cathode. Platinum electrodes are commonly used as the
cathodes.

There have been at least 34 different reactions reported to occur in an
oxygen plasma, however, the vast majority of these are inconsequential
with respect to the formation of active species. Furthermore, many of the
potentially active species are sufficiently short lived that it is unlikely that
they make a significant contribution. The primary active species within the
oxygen plasma are undoubtedly O" and O7*. Both being produced in near
equal quantities, although only the former is relevant to plasma
aniodization. While these species may be active with respect to surface
oxidation, it is more likely that an electron transfer occurs from the
semiconductor surface yields activated oxygen species, which are the actual
reactants in the oxidation of the silicon.

The significant advatage of plasma processes is that while the electron
temperature of the ionized oxygen gas is in excess of 10,000 K, the thermal
temperatures required are significantly lower than required for the high
pressure method, i.e., < 600 °C. The advantages of the lower reaction
temperatures include: the minimization of dopant diffusion and the
impediment of the generation of defects. Despite these advantages there are
two primary disadvantages of any plasma based process. First, the high
electric fields present during the processes cause damage to the resultant
oxide, in particular, a high density of interface traps often result. However,
post annealing may improve film quality. Second, the growth rates of
plasma oxidation are low, typically 1000 A/h. This growth rate is increased
by about a factor of 10 for plasma aniodization, and further improvements
are observed if 1 - 3% chlorine is added to the oxygen source.

Masking

A selective mask against the diffusion of dopant atoms at high temperatures
can be found in a silicon dioxide layer, which can prove to be very useful in
integrated circuit processing. A predeposition of dopant by ion

implantation, chemical diffusion, or spin-on techniques typically results in a
dopant source at or near the surface of the oxide. During the initial high-
temperature step, diffusion in the oxide must be slow enough with respect
to diffusion in the silicon that the dopants do not diffuse through the oxide
in the masked region and reach the silicon surface. The required thickness
may be determined by experimentally measuring, at a particular
temperature and time, the oxide thickness necessary to prevent the inversion
of a lightly doped silicon substrate of opposite conductivity. To this is then
added a safety factor, with typical total values ranging from 0.5 to 0.7 mm.
The impurity masking properties result when the oxide is partially
converted into a silica impurity oxide "glass" phase, and prevents the
impurities from reaching the SiO,-Si interface.

Bibliography

e M. M. Atalla, in Properties of Elemental and Compound
Semiconductors, Ed. H. Gatos, Interscience: New York (1960).

e S. K. Ghandhi, VLSI Fabrication Principles, Silicon and Gallium
Arsenide, Wiley, Chichester, 2nd Ed. (1994).

e S.M. Sze, Physics of Semiconductor Devices, 2nd Edition, John Wiley
& Sons, New York (1981).

e D.L. Lile, Solid State Electron., 1978, 21, 1199.

e W.E. Spicer, P. W. Chye, P. R. Skeath, and C. Y. Su, I. Lindau, J. Vac.
Sci. Technol., 1979, 16, 1422.

e V.Q. Ho and T. Sugano, IEEE Trans. Electron Devices, 1980, ED-27,
1436.

e J. R. Hollanhan and A. T. Bells, Techniques and Applications of
Plasma Chemistry, Wiley, New York (1974).

e R. P.H. Chang and A. K. Sinha, Appl. Phys. Lett., 1976, 29, 56.

Photolithography

Note:This module is based upon the Connexions module entitled
Photolithography by Bill Wilson.

Actually, implants (especially for moats) are usually done at a sufficiently
high energy so that the dopant (phosphorus) is already pretty far into the
substrate (often several microns or so), even before the diffusion starts. The
anneal/diffusion moves the impurities into the wafer a bit more, and as we
Shall see also makes the n-region grow larger.

"The n-region"! We have not said a thing about how we make our moat in
only certain areas of the wafer. From the description we have so far, is
seems we have simply built an n-type layer over the whole surface of the
wafer. This would be bad! We need to come up with some kind of
"window" to only permit the implanting impurities to enter the silicon wafer
where we want them and not elsewhere. We will do this by constructing an
implantation "barrier".

To do this, the first thing we do is grow a layer of silicon dioxide over the
entire surface of the wafer. We talked about oxide growth when we were
discussing MOSFETs but let's go into a little more detail. You can grow
oxide in either a dry oxygen atmosphere, or in a an atmosphere which
contains water vapor, or steam. In [link], we show oxide thickness as a
function of time for growth with steam. Dry Oz does not behave too much
differently, the rate is just somewhat slower.

Xox oxide thickness (um)

time (minutes)

A plot of oxide thickness as a
function of time.

On top of the oxide, we are now going to deposit yet another material. This
is silicon nitride, Si3N, or just plain "nitride" as it is usually called. Silicon
nitride is deposited through a method called chemical vapor deposition or
"CVD". The usual technique is to react dichlorosilane and ammonia in a hot
walled low pressure chemical vapor deposition system (LPCVD). The
reaction is:

Equation:

3 SiH,Cl, + 1ONH, > Si;N, + 6NH,Cl + 6H,

Silicon nitride is a good barrier for impurities, oxygen and other things
which do not want to get into the wafer. Take a look at [link] and see what
we have so far. A word about scale and dimensions. The silicon wafer is
about 250 pm thick (about 0.01") since it has to be strong enough not to
break as it is being handled. The two deposited layers are each about 1 pm
thick, so they should actually be drawn as lines thinner than the other lines
in the figure. This would obviously make the whole idea of a sketch
ridiculous, so we will leave things distorted as they are, keeping in mind
that the deposited and diffused layers are actually much thinner than the rest

of wafer, which really does not do anything except support the active
Circuits up on top.

CDSSSLSISSSSYSSSSSSSSSSSSSSSYSSSSSSSYSSA LLU
RNANANSNANAANAANANANAANASANAANNANNANANAS 54 (0 [7]

silicon

Initial wafer
configuration.

Now what we want to do is remove part of the nitride, so we can make our
n-well, but not put in phosphorous where do not want it. We do this with a
processes called photolithography and etching respectively. First thing we
do is coat the wafer with yet another layer of material. This is a liquid
called photoresist and it is applied through a process called spin-coating.
The wafer is put on a vacuum chuck, and a layer of liquid photoresist is
sprayed uncap of the wafer. The chuck is then spun rapidly, getting to
several thousand RPM in a small fraction of a second. Centrifugal force
causes the resist to spread out uniformly across the wafer surface. The
solvent for the photoresist is quite volatile and so the layer of photoresist
dries while the wafer is still spinning, resulting in a thin, uniform coating
across the wafer [link].

photresist
TOOT nitrides

AAAI AAA KAKA KA AAKAAK ASAI KAAARAAAASALASA
Laat haaaaa OXIGe

silicon

After the photoresist is
spun on.

The name "photoresist" gives some clue as to what this stuff is. Basically,
photoresist is a polymer mixed with some kind of light sensitizing
compound. In positive photoresist, wherever light strikes it, the polymer is
weakened, and it can be more easily removed with a solvent during the
development process. Conversely, negative photoresist is strengthened
when it is illuminated with light, and is more resistant to the solvent than is
the unilluminated photoresist. Positive resist is so-called because the image
of the developed photoresist on the wafer looks just like the mask that was
used to create it. Negative photoresist makes an image which is the opposite
of what the mask looks like.

We have to come up with some way of selectively illuminating certain
portions of the photoresist. Anyone who has ever seen a projector know
how we can do this. But, since we want to make small things, not big ones,
we will change around our projector so that it makes a smaller image,
instead of a bigger one. The instrument that projects the light onto the
photoresist on the wafer is called a projection printer or stepper [link].

Fi \. Light
rd ‘

Lens

, Mask or
‘” Reticle

\ ' ’
/
' ‘ if
1 ‘
.

ate Projection
Lens

“Scan directions

A schematic of a stepper
configuration.

As shown in [link], the stepper consists of several parts. There is a light
source (usually a mercury vapor lamp, although ultra-violet excimer lasers
are also starting to come into use), a condenser lens to image the light
source on the mask or reticle. The mask contains an image of the pattern we
are trying the place on the wafer. The projection lens then makes a reduced
(usually 5x) image of the mask on the wafer. Because it would be far too
costly, if not just plain impossible, to project onto the whole wafer all at
once, only a small selected area is printed at one time. Then the wafer is
scanned or stepped into a new position, and the image is printed again. If
previous patterns have already been formed on the wafer, TV cameras, with
artificial intelligence algorithms are used to align the current image with the
previously formed features. The stepper moves the whole surface of the
wafer under the lens, until the wafer is completely covered with the desired
pattern. A stepper is one of the most important pieces of equipment in the
whole IC fab however, since it determines what the minimum feature size
on the circuit will be.

After exposure, the photoresist is placed in a suitable solvent, and
"developed". Suppose for our example the structure shown in [link] is what
results from the photolithographic step.

photresist
eee nitride

LLU U
IIS SISA SATO OA AAR DAA ORRRAPDIDS
DSSS enbaaaaaaaal OXI

silicon

After photoresist
exposure and
development.

The pattern that was used in the photolithographic (PL) step exposed half of
our area to light, and so the photoresist (PR) in that region was removed
upon development. The wafer is now immersed in a hydrofluoric acid (HF)

solution. HF acid etches silicon nitride quite rapidly, but does not etch
silicon dioxide nearly as fast, so after the etch we have what we see in

After the nitride etch step.

We now take our wafer, put it in the ion implanter and subject it to a "blast"
of phosphorus ions [link].
P P P P,P, PP PPP P

[VLLPLLMIIYOL FLOM preteen
POP

LSS SSS bapa hb Sabb OXide

OS ALBA EVR

silicon

Implanting phospohrus.

The ions go right through the oxide layer on the RHS, but stick in the
resist/nitride layer on the LHS of our structure.

Optical Issues in Photolithography

Note:This module was developed as part of the Rice University course
CHEM-496: Chemistry of Electronic Materials. This module was prepared
with the assistance of Zane Ball.

Introduction

Photolithography is one of the most important technology in the production
of advanced integrated circuits. It is through photolithography that
semiconductor surfaces are patterned and the circuits formed. In order to
make extremely small features, on the order of the wavelength of the light,
advanced optical techniques are used to transfer a pattern from a mask onto
the surface. A polymeric film or resist, is modified by the light and records
the information in a process not dissimilar to ordinary photography.

An illustration of the photolithographic process is shown in [link]. The
process follows the following basic steps:

The wafer is spin coated with resist to form a uniform ~1 pm thin film of
resist on the surface.

The wafer is exposed with ultraviolet light through a mask which contains
the desired pattern. In the simplest processes the mask is simply placed over
the wafer, but advanced sub-micron technologies require the pattern to
imaged through a complex optical system.

The photoresist is developed and the irradiated area is washed away
(positive resist) or the unirradiated area is washed away (negative resist).
Processing (etching, deposition etc.)

Remaining resist is stripped.

UV

me FAL LY | tes

masking film

: : photoresist
(1037, 504) (i) coating
with mask
hotoresist alignment
P. &
——_ >

(ii) sofbake

(i) exposure
(ii) postbake
(ili) development

stripping etching
<—___ <—

Steps in optical printing using photolithography.

In addition to being possibly the most important semiconductor process
step, photolithography is also the most expensive technology in
semiconductor manufacturing. This expense is the result of two
considerations:

1. The optics in photolithography tools are expensive where a single lens
can cost a $1 million or more

2. Each chip (often referred to as a "dye") must be exposed individually
unlike other semiconductor processes such as CVD where an entire
wafer can be processed at a time or oxidation processes where many
wafers can be processed simultaneously.

This means that not only are photolithography machines the most expensive

of semiconductor processing equipment, but more of them are needed in
order to maintain throughput.

Optical issues in photolithography

The critical dimension and depth of focus

A semiconductor process technology is often described by a characteristic
length known as the critical dimension. The critical dimension (CD) is the
smallest feature that needs to be patterned on the surface. The exact
definition varies from process to process but is often the channel length of
the smallest transistor (typical of a memory chip) or the width of the
smallest metal interconnection line (logic chips). This critical dimension is
defined by the photolithographic process and is perhaps the most important
figure of merit in the manufacture of integrated circuits. Making the critical
dimension smaller is the primary focus of improving semiconductor
technology for the following reasons:

1. Making the CD smaller dramatically increases the number of devices
per unit area and this increase goes with the square of the CD (i.e., a
reduction in CD by a factor of 2 generates 4 times the number of
devices).

2. Making the CD smaller of a device already in production will make a
smaller chip. This means that the number of chips per wafer increases
dramatically, and since costs generally scale with the number of wafers
and not the number of chips to a wafer, costs are dramatically reduced.

3. Smaller devices are faster.

Therefore, improvements in lithography technology translate directly into
better, faster, more complex circuits at lower cost.

Having established the importance of the critical dimension it is important
to understand what features of a photolithography system impact. The
theory behind projection lithography is very well known, dating from the
original analysis of the microscope by Abbe. It is, in fact, the Abbe sine
condition that dictates the critical dimension:

Equation:

Xr
CD Coherent — 0. :
nsin(@)

Xr
CD Incoherent — 0. :
nsin(@)

where the two expressions refer to the limit of a purely coherent
illuminating source and purely incoherent source respectively, and A is the
vacuum wavelength of the illuminating light source, n the index of
refraction of the objective lens, and © refers to the angle between the axis
of the lens and the line from the back focal point to the aperture of the
entrance of the lens. The quantity in the denominator, nsin(@) is referred to
as the numerical aperture or NA. As the degree of coherence can be
adjusted in a lithography system, the critical dimension is usually written
more generally as:

Equation:

my

meee sin(@)

From this equation, we begin to see what can be done to reduce the critical
dimension of a lithography system:

1. Change the wavelength of the source.
2. Increase the numerical aperture (NA).
3. Reduce k;.

Before we discuss how this is accomplished, we must consider one other
key quantity, the depth of focus or DOF. The depth of focus is the length
along the axis in which a sharp image exists. Naturally a large DOF is
desirable for ease of alignment, since the entire dye must with lie within
this region. In reality, however, the more meaningful constraint is that the
DOF must be thicker than the resist layer so that the entire volume of resist
is exposed and can be developed. Also, if the surface morphology of the
device dictates that the resist to be exposed is not planar, then the DOF

must be large enough so that all features are properly illuminated. Current
resists must be 1 pm in thickness in order to have the necessary etch
resistance, so this can be considered a minimum value for an acceptable
DOF. The depth of focus can also be expressed as a function of numerical
aperture and wavelength:

Equation:

nN

Benne [nsin(@)]

If we desire to minimize the critical dimension simply by making optics of
large numerical aperture that we will simultaneously reduce the depth of
focus and at a much faster rate owing to the dependence on the square of
the numerical aperture.

These two quantities, DOF and CD, provide the direction in lithography and
semiconductor processing as a whole. For example, a design with an
improved surface planarity or a new resist that is effective at smaller
thicknesses would allow for a smaller depth of focus which would in turn
allow for a larger numerical aperture implying a smaller critical dimension.
The resist, the source wavelength, and the optical delivery system all affect
the critical dimension and that further refinements require a multifaceted
approach to improving lithography systems. What also must be realized is
that, as far as the optical system is concemed, virtually all that can be done
with conventional optics has been done and that fundamental restraints on
k, have been reached.

Wavefront engineering

One way to get around the fundamental limitations of an imaging system
illustrated in [link] is through one of a variety of techniques often termed
wavefront engineering. Here, not only is the amplitude mapped from the
object plane to the image plane, but the phase structure of the light going
through the mask is manipulated to improve the contrast and allow for
effective values of k,; lower than the theoretical minimum for uniform

illumination. The most important example of these techniques is the phase
shift mask or PSM. Here the mask consists of two types of areas, those that
allow light to pass through unaffected and some regions where the
amplitude of the light is unaffected but its phase is shifted. The resulting
electric fields will then sum to zero in some places where use of an ordinary
mask would have resulted in a positive intensity.

There are many problems with the practical introduction of various phase
shifting techniques. Construction of masks with phase shifting elements
(usually a thin layer of PMMA) is difficult and expensive. Mask damage,
already a key problem in conventional production techniques, becomes an
even greater issue as traditional mask repair techniques can no longer be
used. Also identifying errors in a mask is made more difficult by the odd
design.

Interaction with resists

The ultimate resolution of a photolithographic process is not dependent on
optics alone, but also on the interaction with the resist. One of the key
concerns, particularly as wavelengths of sources become shorter, is the
ability of the source light to penetrate the resist film. Many polymers absorb
strongly in the UV which can limit the interaction to the surface. In such a
case only a thin layer of the polymer is exposed and the pattern may not be
fully uncovered during developing. One important property of resist is the
presence of saturable absorption.. Saturable absorbers are those absorption
sites in the polymer that when excited to a higher state remain there for
relatively long periods of time and do not continue to absorb into higher
states. If only saturable absorption is present in a polymer film, then
continued irradiation eventually leads to transparency as all absorption sites
will be saturated. This allows light penetration through the resist film with
full exposure to the substrate surface.

Full penetration of the film leads to a second problem, multiple reflection
interference. This occurs when light which has penetrated the film to the
substrate is then reflected back towards the surface. The result is a standing
wave interference pattern which causes uneven exposure through the film.

The problem becomes more severe as optical limits are approached where
feature size is approximately equal to the wavelength of the light source
meaning such standing waves are the same size as the irradiated features. In
the most advanced lithography techniques such as 248 nm lithography with
excimer lasers, a special anti-reflectance coating must be laid down before
the resist is deposited. Development of an AR coating that has no adverse
effects during the exposure and development process is difficult.

One completely new approach to photolithography resists are top-surface-
imaged resists or TSI resists. These processes do not require light
penetration through the whole volume of resist. In a TSI resist, a silyl amine
is selectively in-diffused from the gas phase into a phenolic polymer in
response to the laser irradiation. This diffusion process creates a silyl ether,
and development takes place in the form of an oxygen plasma etch,
sometimes termed 'dry developing’. Depth of focus limitations are thus
avoided as exposure is necessary only at the surface of the resist layer, and
the resolution of the etching process determines the final resist profile. Such
a technique has tremendous advantages, particularly as source wavelengths
become shorter and transparent polymers more rare. Such as resist has a
clear optical advantage as well since the image need only be formed at the
surface of the resist layer reducing the DOF needed to 100 nm or less,
allowing for larger numerical aperture lithography systems with smaller
critical dimensions.

Light sources

Current photolithography techniques in production utilize ultraviolet lamps
as the light source. In the most advanced production facilities, 0.35 pm
mercury i-line technology is used. For the next generation of chips such as
64 Mbit DRAMS better performance is necessary and either i-line
technology combined with PSM or a new light source is required. Certainly
for the 256 Mbit generation using 0.25 jam technology, the i-line source is
no longer adequate. The apparent successor is the 248 nm KrF laser, which
entered the most advanced production facilities in the late 1990s. KrF
technology is often referred to in the literature as Deep UV or DUV
lithography. For further shrinkage to 0.18 am technology, the ArF excimer

laser at 193 nm will likely be used with the transition likely to take place in
the first few years of the next decade.

At critical dimensions lower than 0.18 - 0.1 pm and below, a whole host of
technological problems will need to be overcome in every stage of
manufacturing including photolithography. One likely scheme for future
lithography is to use X-rays where the wavelength of the light is so much
smaller than the feature size such that proximity printing can be used. This
is where the mask is placed close to the surface and an X-ray source is
scanned across using no optics. Common X-ray sources for such techniques
include synchrotron radiation and laser produced plasmas. It has also been
widely suggested that the cost of implementing X-ray or other post-optical
techniques together with the increased cost of every other manufacturing
process step will make improvements beyond 0.1 ym cost prohibitive
where benefits in increased circuit speed and density will be dwarfed by
massive manufacturing cost. It is noted however that such predictions have
been made in the past with regard to other technological barriers.

Bibliography

e M. Born and E. Wolf, Principles of Optics 6th Edition, Pergamon
Press, New York (1980).

e M. Nakase, IEICE Trans. Electron., 1993, E76-C, 26.

e M. Rothschild, A. R. Forte, M. W. Horn, R. R. Kunz, S. C. Palmateer,
and J. H. C. Sedlacek, IEEE J. Selected Topics in Quantum
Electronics, 1995, 1, 916.

Composition and Photochemical Mechanisms of Photoresists

Note:This module was developed as part of the Rice University course
CHEM-496: Chemistry of Electronic Materials. This module was prepared
with the assistance of Angela Cindy Wei.

Photolithography

In photolithography, a pattern may be transferred onto a photoresist film by
exposing the photoresist to light through a mask of the pattern. In the
semiconductor industry, the photolithographic procedure includes the
following steps as illustrated in [link]: coating a base material with
photoresist, exposing the resist through a mask to light, developing the
resist, etching the exposed areas of the base, and stripping the remaining
resist off.

UV

mg SALLY yt

masking film

(SiOp, SizN4) photoresist

(i) coating

with mask
photoresist alignment
(ii) sofbake

(i) exposure
(ii) postbake
(iii) development

stripping etching
<m — <m —

Steps in optical printing using photolithography.

Upon exposure to light, the photoresist may become more or less soluble
depending on the chemical properties of the particular resist material. The
photochemical reactions include chain scission, cross-linking, and the
rearrangement of molecules. If the exposed areas of the photoresist become
more soluble, then it is a positive resist; conversely, if the exposed resist
becomes less soluble, then it is a negative resist. In developing the
photoresist, the more soluble material is removed leaving a positive or a
negative image of the mask pattern.

Photoresist

Photoresists were initially developed for the printing industry. In the 1920s,
the application of photoresists spread to the printed circuit board industry.
Photoresists for semiconductor use were first developed in the 1950s;
Kodak developed commercial negative photoresists and shortly after,
Shipley developed a line of positive resists. Several other companies have
entered the market since that time in hopes of manufacturing resist products
which meet the increasing demands of the semiconductor industry:
narrower line widths, fewer defects, and higher production rates.

Photoresist composition

Several functional requirements must be met for a photoresist to be used in
the semiconductor industry. Photoresist polymers must be soluble for easy
deposition onto a substrate by spin-coating. Good photoresist-substrate
adhesion properties are required to minimize undercutting, to maintain edge
acuity, and to control the feature sizes. The photoresist must be chemically
resistant to whichever etchants are to be used. Sensitivity of the photoresist
to a particular light source is essential to the functionality of a photoresist.
The speed at which chemical changes occur in a photoresist is its contrast.
The contrast of a resist is dependent on the molecular weight distribution of

the polymers: a broad molecular weight distribution results in a low contrast
resist. High contrast resists produce higher resolution images.

The four basic components of a photoresist are the polymer, the solvent,
sensitizers, and other additives. The role of the polymer is to either
polymerize or photosolubilize when exposed to light. Solvents allow the
photoresist to be applied by spin-coating. The sensitizers control the
photochemical reactions and additives may be used to facilitate processing
or to enhance material properties. Photochemical changes to polymers are
essential to the functionality of a photoresist. Polymers are composed
primarily of carbon, hydrogen, and oxygen-based molecules arranged in a
repeated pattern. Negative photoresists are based on polyisopreme
polymers; negative resist polymers are not chemically bonded to each other,
but upon exposure to light, the polymers crosslink, or polymerize. Positive
photoresists are formulated from phenol-formaldehyde novolak resins; the
positive resist polymers are relatively insoluble, but upon exposure to light,
the polymers undergo photosolubilization.

Solvents are required to make the photoresist a liquid, which allows the
resist to be spun onto a substrate. The solvents used in negative photoresists
are non-polar organic solvents such as toluene, xylene, and halogenated
aliphatic hydrocarbons. In positive resists, a variety of organic solvents such
as ethyl cellosolve acetate, ethoxyethyl acetate, diglyme, or cyclohexanone
may be used.

Photosensitizers are used to control or cause polymer reactions resulting in
the photosolubilization or crosslinking of the polymer. The sensitizers may
also be used to broaden or narrow the wavelength response of the
photoresist. Bisazide sensitizers are used in negative photoresists while
positive photoresists utilize diazonaphthoquinones. One measure of
photosensitizers is their quantum efficiencies, the fraction of photons which
result in photochemical reactions; the quantum efficiency of positive
diazonaphthoquinone photoresist sensitizers has been measured to be 0.2 -
0.3 and the quantum efficiency of negative bis-arylazide sensitizers is in the
range of 0.5 - 1.0.

Additives are also introduced into photoresists depending on the specific
needs of the application. Additives may be used to increase photon

absorption or to control light within the resist film. Adhesion promoters
such as hexamethyldisilazane and additives to improve substrate coating are
also commonly used.

Negative photoresist chemistry

The matrix resin material used in the formulation of these (negative) resists
is a synthetic rubber obtained by a Ziegler-Natta polymerization of isoprene
which results in the formation of poly(cis-isoprene). Acid-catalyzation of
poly(cis-isoprene) produces a partially cyclized polymer material; the
cyclized polymer has a higher glass transition temperature, better structural
properties, and higher density. On the average, microelectronic resist
polyisoprenes contain 1-3 rings per cyclic unit, with 5-20% unreacted
isoprene units remaining’. The resultant material is extremely soluble in
non-polar, organic solvents including toluene, xylene, and halogenated
aliphatic hydrocarbons.

The condensation of para-azidobenzaldehyde with a substituted
cyclohexanone produces bis-arylazide sensitizers. To maximize the
absorption of a particular light source, the absorbance spectrum of the
photoresist may be shifted by making structural modifications to the
sensitizers; for example, by using substituted benzaldehydes, the absorption
peak may be shifted to longer wavelengths. A typical bisazide-cyclized
polyisoprene photoresist formulation may contain 97 parts cyclized
polyisoprene to 3 parts bisazide in a (10 wt%) xylene solvent.

All negative photoresists function by cross-linking a chemically reactive
polymer via a photosensitive agent that initiates the chemical cross-linking
reaction. In the bisazide-cyclized polyisoprene resists, the absorption of
photons by the photosensitive bisazide in the photoresist results in an
insoluble crosslinked polymer. Upon exposure to light, the bisazide
sensitizers decompose into nitrogen and highly reactive chemical
intermediates, called nitrenes [link]. The nitrines react to produce polymer
linkages and three-dimensional cross-linked structures that are less soluble
in the developer solution.

Equation:

CH;
Ng
hv
SSS RNS tN;
N
O

Positive photoresist chemistry

Positive photoresist materials originally developed for the printing industry
have found use in the semiconductor industry. The commonly used novolac
resins (phenol-formaldehyde copolymer) and (photosensitive) diazoquinone
both were products of the printing industry.

The novolak resin is a copolymer of a phenol and formaldehyde ([link]).
Novolak resins are soluble in common organic solvents (including ethyl
cellosolve acetate and diglyme) and aqueous base solutions. Commercial
resists usually contain meta-cresol resins formed by the acid-catalyzed
condensation of meta-cresol and formaldehyde.

OH OH

Structure of a novolak resin.

The positive photoresist sensitizers are substituted diazonaphthoquinones.
The choice of substituents affects the solubility and the absorption

characteristics of the sensitizers. Common substituents are aryl sulfonates.
The diazoquinones are formed by a reaction of diazonaphthoquinone
sulfonyl chloride with an alcohol to form sulfonate ester; the sensitizers are
then incorporated into the resist via a carrier or bonded to the resin. The
sensitizer acts as a dissolution inhibitor for the novalac resin and is base-
insoluble. The positive photoresist is formulated from a novolac resin, a
diazonaphthoquinone sensitizer, and additives dissolved in a 20 - 40 wt%
organic solvent. In a typical resist, up to 40 wt% of the resist may be the
sensitizer.

The photochemical reaction of quinonediazide is illustrated in [link]. Upon
absorption of a photon, the quinonediazide decomposes through Wolff
rearrangement, specifically a Sus reaction, and produces gaseous nitrogen
as a by-product. In the presence of water, the decomposition product forms
an indene carboxylic acid, which is base-soluble. However, the formation of
acid may not be the reason for increased solubility; the release of nitrogen
gas produces a porous structure through which the developer may readily
diffuse, resulting in increased solubility.

Equation:
O
q OH
Np hv
+ H,O
Base-insoluble Base-soluble
sensitizer photoproduct

Image reversal

By introducing an additive to the novolac resins with diazonaphthaquiones
sensitizers, the resultant photoresist may be used to form a negative image.
A small amount of a basic additive such as monazoline, imidazole, and
triethylamine is mixed into a positive novolac resist. Upon exposure to

light, the diazonaphthaquiones sensitizer forms an indene carboxylic acid.
During the subsequent baking process, the base catalyzes a thermal
decarboxylation, resulting in a substituted indene that is insoluble in
aqueous base. Then, the resist is flood exposed destroying the dissolution
inhibitors remaining in the previously unexposed regions of the resist. The
development of the photoresist in aqueous base results in a negative image
of the mask.

Comparison of positive and negative photoresists

Into the 1970s, negative photoresist processes dominated. The poor
adhesion and the high cost of positive photoresists prevented its widespread
use at the time. As device dimensions grew smaller, the advantages of
positive photoresists, better resolution and pinhole protection, suited the
changing demands of the semiconductor industry and in the 1980s the
positive photoresists came into prominence. A comparison of negative and
positive photoresists is given in [link].

radiation
(UV, e°, X-ray, ions)

mask
—————— as
resist
SiO,
Si-substrate
Develop
Etching
&
Stripping
Positive Resist Negative Resist

A comparison of negative and positive photoresists.

The better resolution of positive resists over negative resists may be
attributed to the swelling and image distortion of negative resists during
development; this prevents the formation of sharp vertical walls of negative
resist. Disadvantages of positive photoresists include a higher cost and
lower sensitivity.

Positive photoresists have become the industry choice over negative
photoresists. Negative photoresists have much poorer resolution and the
positive photoresists exhibit better etch resistance and better thermal
stability. As optical masking processes are still preferred in the
semiconductor industry, efforts to improve the processes are ongoing.
Currently, researchers are studying various forms of chemical amplification
to increase the photon absorption of photoresists.

Bibliography

W.M. Alvino, Plastics For Electronics, McGraw-Hill, Inc, New York
(1995).

R. W. Blevins, R. C. Daly, and S. R. Turner, in Encyclopedia of
Polymer Science and Engineering, Ed. J. 1. Krocehwitz, Wiley, New
York (1985).

M. J. Bowden, in Materials for Microlithography: Radiation-Sensitive
Polymers, Ed. L. F. Thompson, C. G. Willson, and J. M. J. Frechet,
American Chemical Society Symposium Series No. 266, Washington,
D.C. (1984).

S. J. Moss and A. Ledwith, The Chemistry of the Semiconductor
Industry, Blackie & Son Limited, Glasgow (1987).

E. Reichmanis, F. M. Houlihan, O. Nalamasu, and T. X. Neenan, in
Polymers for Microelectronics, Ed. L. F. Thompson, C. G. Willson,
and S. Tagawa, American Chemical Society Symposium Series, No.
537, Washington, D.C. (1994).

P. van Zant, Microchip Fabrication, 2nd ed., McGraw-Hill Publishing
Company, New York (1990).

C. Grant Willson, in Introduction to Microlithography, 2nd ed., Ed. L.
F. Thompson, C. G. Willson, M. J. Bowden, American Chemical
Society, Washington, D.C. (1983).

Integrated Circuit Well and Gate Creation

Note: This module is based upon the Connexions module entitled
Integrated Circuit Well and Gate Creation by Bill Wilson.

We then remove the remaining resist, and perform an
activation/anneal/diffusion step, also sometimes called the "drive-in". The
purpose of this step is two fold. We want to make the n-tank deep enough so
that we can use it for our p-channel MOS, and we want to build up an
implant barrier so that we can implant into the p-substrate region only. We
introduce oxygen into the reactor during the activation, so that we grow a
thicker oxide over the region where we implanted the phosphorus. The
nitride layer over the p-substrate on the LHS protects that area from any
oxide growth. We then end up with the structure shown in [link].

nitride

After the anneal/drive-in.

Now we strip the remaining nitride. Since the only way we can convert
from p to n is to add a donor concentration which is greater than the
background acceptor concentration, we had to keep the doping in the
substrate fairly light in order to be able to make the n-tank. The lightly
doped p-substrate would have too low a threshold voltage for good n-MOS
transistor operation, so we will do a Vr adjust implant through the thin
oxide on the LHS, with the thick oxide on the RHS blocking the boron from
getting into the n-tank. This is shown in [link], where boron is implanted

into the p-type substrate on the LHS, but is blocked by the thick oxide in

the region over the n-well.
BBB BBB BB

aa ot

Vy adjust implant.

Next, we strip off all the oxide, grow a new thin layer of oxide, and then a
layer of nitride [link]. The oxide layer is grown only because it is bad to
grow Si3N, directly on top of silicon, as the different coefficients of thermal
expansion between the two materials causes damage to the silicon crystal
structure. Also, it turns out to be nearly impossible to remove nitride if it is
deposited directly on to silicon.

sacrificial _
j nitride

SSSsScssiiscccsuaaies
MS OER ]

UI Tse TH]

Strip of the oxide and
grow a new nitride layer.

The nitride is patterned (covered with photoresist, exposed, developed,
etched, and removal of photoresist) to make two areas which are called
"active" [link]. The wafer is then subjected to a high-pressure oxidation step
which grows a thick oxide wherever the nitride was removed. The nitride is
a good barrier for oxygen, so essentially no oxide grows underneath it. The

thick oxide is used to isolate individual transistors, and also to make for an
insulating layer over which conducting patterns can be run. The thick oxide
is called field oxide (or FOX for short) [link].

nitride to define "active" regions

Nitride remaining after
etching.

NZS N
NEE SEIT

et

After growth of the field
oxide (FOX).

Then, the nitride, and some of the oxide are etched off. The oxide is etched
enough so that all of the oxide under the nitride regions is removed, which
will take a little off the field oxide as well. This is because we now want to
grow the gate oxide, which must be very clean and pure [link]. The oxide
under the nitride is sometimes called a sacrificial oxide, because it is
sacrificed in the name of ultra performance.

Ready to grow gate
oxide.

Then the gate oxide is grown, and immediately thereafter, the whole wafer
is covered with polysilicon [link].

Polysilicon deposition
over the gate oxide.

The polysilicon is then patterned to form the two regions which will be our
gates. The wafer is covered once again with photoresist. The resist is
removed over the region that will be the n-channel device, but is left
covering the p-channel device. A little area near the edge of the n-tank is
also uncovered [link]. This will allow us to add some additional phosphorus
into the n-well, so that we can make a contact there, so that the n-well can
be connected to Vag.

Preparing for NMOS
channel/drain implant.

Back into the implanter we go, this time exposing the wafer to phosphorus.
The poly gate, the FOX and the photoresist all block phosphorus from
getting into the wafer, so we make two n-type regions in the p-type
substrate, and we have made our n-channel MOS source/drain regions. We
also add phosphorous to the Vag contact region in the n-well so as the make

sure we get good contact performance there [link].
P PP P

Phosphorus S/D implant.

The formation of the source and drain were performed with a self-aligning
technology. This means that we used the gate structure itself to define
where the two inside edges of the source and drain would be for the
MOSFET. If we had made the source/drain regions before we defined the
gate, and then tried to line the gate up right over the space between them,
we might have gotten something that looks like what is shown in [link].
What's going to be the problem with this transistor? Obviously, if the gate

does not extend all the way to both the source and the drain, then the
channel will not either, and the transistor will never turn on! We could try
making the gate wider, to ensure that it will overlap both active areas, even
if it is slightly misaligned, but then you get a lot of extraneous fringing
capacitance which will significantly slow down the speed of operation of
the transistor [link]. This is bad! The development of the self-aligned gate
technique was one of the big breakthroughs which has propelled us into the
VLSI and ULSI era.

eeu

A representation of a
misaligned gate.

~
\Y pee |
\ mn MT WN

A representation of a
wide gate.

We pull the wafer out of the implanter, and strip off the photoresist. This is
sometimes difficult, because the act of ion implantation can "bake" the
photoresist into a very tough film. Sometimes an rf discharge in an O>
atmosphere is used to "ash" the photoresist, and literally burn it off the
wafer! We now apply some more PR, and this time pattern to have the moat
area, and a substrate contact exposed, for a boron p* implant. This is shown
in [Link].

BBBB

OS aueazuni nun ree
EK Veen f

pee “) | eed
drain Altseot|

p-type source

Boron p-channel S/D
implant.

We are almost done. The next thing we do is remove all the photoresist, and
grow one more layer of oxide, which covers everything, as shown in [link].
We put photoresist over the whole wafer again, and pattern for contact holes
to go through the oxide. We will put contacts for the two drains, and for
each of the sources, make sure that the holes are big enough to also allow us
to connect the source contact to either the p-substrate or the n-moat as is

appropriate [link].

Final oxide growth.

After the contact holes
are etched.

Applying Metallization by Sputtering

Note:This module is adapted from the Connexions module entitled
Applying Metal/Sputtering by Bill Wilson.

We now put the wafer in a sputter deposition system. In the sputter system,
we Coat the entire surface of the wafer with a conductor. An aluminum-
silicon alloy is usually used, although other metals are employed as well.

A sputtering system is shown schematically in [link]. A sputtering system is
a vacuum chamber, which after it is pumped out, is re-filled with a low-
pressure argon gas. A high voltage ionizes the gas, and creates what is
known as the Crookes dark space near the cathode, which in our case,
consists of a metal target made out of the metal we want to deposit. Almost
all of the potential of the high-voltage supply appears across the dark space.
The glow discharge consists of argon ions and electrons which have been
stripped off of them. Since there are about equal number of ions and
electrons, the net charge density is about zero, and hence by Gauss' law, so
is the field.

Target

discharge

Crook's
dark space

Substrates =

A schematic representation of a sputtering
apparatus.

The electric field accelerates the argon atoms which slam into the aluminum
target. There is an exchange of momentum, and an aluminum atom is
ejected from the target ([link]) and heads to the silicon wafer, where it

sticks, and builds up a metal film [link].
Aluminum Target

ELECTRIC FIELD

The sputtering mechanism.

AOAAAAN
: ¥ ps 2

Wafer coated with metal.

If you look at [link], you will note that we have seemingly done something
pretty stupid. We have wired all of the elements of our CMOS inverter
together; but all is not lost. We can do one more photolithographic step, and
pattern and etch the aluminum, so we only have it where we need it. This is
shown in [link].

After interconnect patterning.

Molecular Beam Epitaxy

Note: This module was developed as part of the Rice University course
CHEM-496: Chemistry of Electronic Materials. This module was prepared
with the assistance of Sarah Westcott.

Introduction

In the process of epitaxy, a thin layer of material is grown on a substrate.
With respect to crystal growth it applies to the process of growing thin
crystalline layers on a crystal substrate. In epitaxial growth, there is a
precise crystal orientation of the film in relation to the substrate. For
electronic devices, the substrate is a single crystal (usually Si or GaAs) and
therefore so is the epitaxial layer (epilayer). In the most basic form of
molecular beam epitaxy (MBE), the substrate is placed in ultra high
vacuum (UHV) and the source materials for the film are evaporated from
elemental sources. The evaporated molecules or atoms flow as a beam,
striking the substrate, where they are adsorbed on the surface. Once on the
surface, the atoms move by surface diffusion until they reach a
thermodynamically favorable location to bond to the substrate. Molecules
will dissociate to atomic form during diffusion or at a favorable site. [link]
illustrates the processes that can occur on the surface. Because the atoms
require time for surface diffusion, the quality of the film will be better with
slower growth. Typically growth rates of about 1 monolayer per second
provide sufficiently high quality.

Deposition

Downward
funneling

Nucleation

Schematic illustration of processes on
growing surface during MBE. Adsorption of
atoms on the surface, surface diffusion of
atoms, formation of crystalline lattice,
desorption of particles from the surface.

A typical MBE chamber is shown in [link]. The substrate is chemically
washed and then put into a loading chamber where it is further cleaned
using argon ion bombardment followed by annealing. This removes the top
layers of the substrate, which is usually an undesired oxide which grew in
air and contains impurities. The annealing heals any damage caused by the
bombardment. The substrate then enters the growth chamber via the sample
exchange load lock. It is secured on a molybdenum holder either
mechanically or with melted indium or gallium which hold the substrate by
surface tension.

Sample heating
and rotation

RHE Chamber cooled
( ae, by liq N2

> Effusion cells

Shutter

Fluorescent screen
(RHEED)

The MBE growth chamber.

Each effusion cell (see [link]) is a source of one element in the film. The
effusion cell, also called a Knudsen cell, contains the elemental form in
very high purity (greater than 99.99999% for Ga and As). The cell is heated
to encourage evaporation. For GaAs growth, the temperature is typically
controlled for a vapor pressure of 10° to 10°? Torr inside the effusion cell,
which results in a transport of about 10!° molecules/cm? to the substrate
when the shutter for that cell is opened. The shape and size of the opening
in the cell is optimized for an even distribution of particles on the substrate.
Due to the relatively low concentration of molecules, they typically do not
interact with other molecules in the beam during the 5 - 30 cm journey to
the substrate. The substrate is usually rotated, at a few rpm, to further even
the distribution.

Because MBE takes place in UHV and has relatively low pressure of
residual gas at the surface, analysis techniques such as reflection high
energy diffraction and ellipsometry can be used during growth, both to
study and control the growth process. The UHV environment also allows
pre or post growth analysis techniques such as Auger spectroscopy.

Elemental and molecular sources

The effusion cell is used for the majority of MBE growth. All materials
used in the cell are carefully chosen to be noninteracting with the element
being evaporated. For example, the crucible is pyrolitic boron nitride.
However, it has disadvantages, such as:

e The evaporated species may be molecular, rather than monomeric,
which will require further dissocation at the surface.

e When the shutter is opened, the heat loss from the cell results in a
transient in the beam flux which last for several minutes and cause
variations of up to 50%.

e The growth chamber must be opened up to replace the solid sources.

Cracker cells are used to improve the ratio of monomeric to molecular (or at
least dimeric to tetrameric) particles from the source. The cracker cell,
placed so that the beam passes through it after the effusion cell, is
maintained at a high temperature (and sometimes high pressure) to
encourage dissociation. The dissociation process generally requires a
catalyst and the best catalysts for a given species have been studied.

Some elements, such as silicon, have low enough vapor pressure that more
direct heating techniques such as electron bombardment or laser radiation
heating are used. The electron beam is bent using electromagnetic focusing
to prevent any impurities in the electron source from contaminating the
silicon to be used in MBE. Because the heat is concentrated on the surface
to be evaporated, interactions with and contamination from the crucible
walls is reduced. In addition, this design does not require a shutter, so there
is no problem with transients. Modulation of the beam can produce very
sharp interfaces on the substrate. In laser radiation heating, the electron
beam is replaced by a laser beam. The advantages of localized heating and
rapid modulation are also maintained without having to worry about
contamination from the electron source or stray electrons.

Some of the II-VI (12-16) compounds have such high vapor pressure that a
Knudson cell cannot be used. For example, the mercury source must be
kept cooler than the substrate to keep the vapor pressure low enough to be

feasible. The Hg source must also be sealed off from the growth chamber to
allow the chamber to be pumped down.

Two other methods of obtaining the elements for use in epitaxy are gas-
source epitaxy and chemical beam epitaxy (CBE). Both of these methods
use gas sources, but they are distinguished by the use of elemental beams in
gas source epitaxy, while organometallic beams are used in CBE. For the
example of III-V (13-15) semiconductors, in gas epitaxy, the group III
material may come from an effusion cell while the group V material is the
hydride, such as AsH3 or PH3, which is cracked before entering the growth
chamber. In CBE, the group V material is an organometallic, such as
triethylgallium [Ga(C>Hs)3] or trimethylaluminum [AI(CH3)3], which
adsorbs on the surface, where it dissociates.

The gas sources have several advantages. Gas lines can be run into the
chamber, which allows the supply to be replenished without opening the
chamber. When making alloys, such as Al,Ga,_,As, the gases can be
premixed for the correct stochiometry or even have their composition
gradually changed for making graded structures. For abrupt structures, it is
necessary to be able to switch the gas lines with speeds of 1 second or less.
However, the gas lines increase the complexity of the process and can be
hard on the pumping system.

Substrate choice and preparation

Materials can be grown on substrates of different structure, orientation, and
chemistry. In deciding which materials can be grown on a particular
substrate, a primary consideration was expected to be lattice mismatch, i.e.,
differences in spacing between atoms. However, while lattice mismatch can
cause strain in the grown layer, considerable accommodation between
materials of different sizes can take place during growth. A greater source
of strain can be differences in thermal expansion characteristics because the
layer is grown at high temperature. On cooling to room temperature,
dislocation defects can be formed at the interface or in severe cases, the
device may break. Chemical considerations, such as whether the layer's
elements will dissolve in the substrate or form compounds with the
substrate, must also be considered.

Different orientations of the substrate can also affect growth. Close-packed
planes have the lowest surface energy, which allows atoms to desorb from
the surface, resulting in slower growth rates. Growth is favored where
bonds can be made in several directions at the same time. Therefore, the
substrate is often cut off-axis by a 2 - 4° to provide a rougher growth
surface. For compound semiconductors, some orientations result in the
number of loose bonds changing between layers. This results in changes of
surface energy with each layer, which slows growth down.

The greatest cause of defects in the epitaxial layer is defects on the
substrate's surface. In general, any dislocations on the substrate are
replicated or even multiplied in the epitaxial growth, which is what makes
the cleaning of the substrate so important.

Materials grown

MBE is commercially used primarily for GaAs devices. This is partly
because the high speed microwave devices made from GaAs required the
superior electrical quality of epitaxial layers. Taking place at lower
temperature and under better controlled conditions, MBE generally results
in layers of better quality than melt-grown.

From solid Ga and As sources, elemental Ga and tetrameric As, are
evaporated. For a GaAs substrate, the Ga flux has a sticking coefficient very
close to 1 (almost certain to adsorb). The As is much less likely to adsorb,
so an excess is usually supplied. Cracker cells are often used on the As, in
order to obtain As» instead, which results in faster growth and more
efficient utilization of the source beam.

For nominally undoped GaAs grown by MBE, the residual impurities are
usually carbon, from substrate contamination and residual gas after the
growth chamber is pumped down, and sulphur, from the As source. The
most common surface defects are "oval" defects, which seem to form when
Ga manages to form metallic droplets during the growth process, which can
particularly occur if the substrate was not cleaned properly. These defects
can also be reduced by carefully controlling the Ga flux.

During MBE growth, dopants can be introduced by having a separate
effusion cell or gas source for each dopant. To achieve a desired dopant
concentration in the film, not only must the rate of dopants striking the
substrate be controlled, but the characteristics of how the dopant behaves on
the surface must be known. Low-vapor pressure dopants tend to desorb
from the surface and their behavior is very temperature dependent and so
are avoided when possible. Slow diffusing dopants adsorb to surface sites
and are eventually covered as more GaAs is grown. Their incorporation
depends linearly on the partial pressure of the dopant present in the growth
chamber. This is the behavior exhibited by most n-type dopants in GaAs
and most dopants of both types in Si. If the dopant, like most p-type GaAs
dopants, is able to diffuse through the surface of the substrate into the
crystal below, then there will be higher incorporation efficiency, which will
depend on the square root of the dopant partial pressure for reasonable
concentrations. Due to increasing lattice strain, all dopants will saturate at
very high concentrations. They may also tend to form clusters. Dopant
behavior depends on many factors and is actively studied.

The growth of GaAs epitaxial layers on silicon substrates has also been
investigated. Silicon substrates are grown in larger wafers, have better
thermal conductivity allowing more devices/chip to be grown on them, and
are cheaper. However, because Si is nonpolar and GaAs is polar, the GaAs
tends to form islands on the surface with different phase (what should be a
Ga site based on a neighboring domain's pattern will actually be an As site).
There is also a fairly large lattice mismatch, leading to may dislocations.
However, FETs, LEDs, and lasers have all been made in laboratories.

Many devices require abrupt junctions between layers of different materials.
One group, studying how to make high quality, abrupt GaAs and AlAs
layers, found that rapid movement of the Ga or Al on the surface was
required. This migration was enhanced at high temperatures, but
unfortunately, diffusion into the substrate also increased. However, they
also discovered that migration of Ga or Al increased if the As supply was
turned off. By alternating the Ga and As supplies, the Ga was able to reach
the substrate and migrate to provide more even monolayer coverage before
the As atoms arrived to react.

Besides GaAs, most other III-V semiconductors have also been grown
using MBE. Structures involving very thin layers (only a few atomic layers
thick), often called superlattices or strained superlattices if there is a large
lattice mismatch, are routinely grown. Because different materials have
different energy levels for electrons and holes, it is possible to trap carriers
in one of these thin layers, forming a quantum well. This type of
confinement structure is particularly popular for LEDs or lasers, including
blue light lasers. The strained superlattice structure actually shifts and splits
the energy levels of the materials in some cases making devices possible for
such applications as infrared light detection, which requires very small band

gaps.

Thin films of many other materials have also been grown using MBE
methods. Silicon technology has cheaper methods of growth and so Si
layers are not very popular. However, possible devices made of Si-Ge alloys
have been grown. The II-VI compounds, have also been grown. Magnetic
materials, such as Co-Pt and Fe-Pt alloys, have been grown in the hopes of
providing better magnetic storage.

Analysis techniques

The most popular in-situ analysis technique for MBE-grown layers is
reflection high energy diffraction (RHEED), see [link]. Electrons of energy
5 - 40 keV are directed towards the sample. They reflect from the surface at
a very small angle (less than 3°) and are directed onto a screen. These
electrons interact with only the top few atomic layers and thus provide
information about the surface. [link] shows a typical pattern on the screen
for electrons reflected from a smooth surface, in which constructive
interference between some of the electrons reflected from the lattice
structure results in lines. If the surface is rough, spots will appear on the
screen, also. By looking at the total intensity of the reflected electron
pattern, an idea of the number of monolayers deposited and how epilayers
grow can be obtained. The island-type growth shown in this figure is an
area of intense interest. These oscillations in intensity are gradually damped
as more layers are grown, because the overall roughness of the surface
increases.

Incident
electron beam

Schematic illustrating the formation of a
RHEED pattern.

RHEED diffraction pattern of a GaAs surface. Adapted
from images by the MBE Laboratory in the Institute of
Physics of the ASCR
(http://www. fzu.cz/departments/surfaces/mbe/index.html)

Phase locked epitaxy takes advantage of the patterns of the oscillations to
grow very abrupt layers. By sending the oscillation information to a
computer, it can decide when to open or close the shutters of the effusion
cell based on the location in the oscillation cycle. This technique self-
adjusts for fluctuations in beam flux when the shutters are opened and can
grow very abrupt layers.

Another analysis technique that can be used to study surface smoothness
during growth is ellipsometry. Polarized laser light is reflected from the
surface at a small angle. The polarization of the light changes, depending on
the roughness of the surface.

Improved growth characteristics also require that the actual flux from the
sources is measured. This is typically done with an ion gauge flux monitor,
which is either used to measure residual beam that misses the substrate or is
moved into the beam path for calibration when a new source is used.
Because of the importance of clean substrate surfaces for low-defect
growth, Auger spectroscopy is used following cleaning by sputtering.
Auger spectroscopy takes place by ionizing an inner shell electron from an
atom. When an outer shell electron then deexcites to the inner shell, the
energy released can prompt the emission of another outer shell electron.
The energy at which this occurs is characteristic of the atom involved and
the signal can be used to detect impurities as small as 0.1%.

Bibliography

e K. J. Bachmann, The Materials Science of Microelectronics, VCH
(1995).

e S. K. Ghandhi, VLSI Fabrication Principles: Silicon and Gallium
Arsenide, 2nd Edition, Wiley-Interscience, NY (1994).

e M. A. Herman and H. Sitter, Molecular Beam Epitaxy: Fundamentals
and Current Status, Springer-Verlag (1989).

e Y. Horikoshi, M. Kawashima, and H. Yamaguchi, Jpn. J. Appl. Phys.,
1986, 25, L868.

e J. H. McFee, B. I. Miller, and K. J. Bachmann, J. Electrochem. Soc.,
1977, 124, 259.

e T. Sakamoto, H. Funabashi, K. Ohta, T. Nakagawa, N. J. Kawai, and T.
Kojima, Jpn. J. Appl. Phys., 1984, 23, L657.
e W. T. Tsang, J. Crystal Growth, 1987, 81, 261.

Atomic Layer Deposition

This module was developed as part of the Rice University course CHEM-
496: Chemistry of Electronic Materials. This module was prepared with the
assistance of Julie A. Francis.

Introduction

The growth of thin films has had dramatic impact on technological
progress. Because of the various properties and functions of these films,
their applications are limitless especially in microelectronics. These layers
can act as superconductors, semiconductors, conductors, insulators,
dielectric, or ferroelectrics. In semiconductor devices, these layers can act
as active layers and dielectric, conducting, or ion barrier layers. Depending
on the type of film material and its applications, various deposition
techniques may be employed. For gas-phase deposition, these include
vacuum evaporation, reactive sputtering, chemical vapor deposition (CVD),
especially metal organic CVD (MOCVD), and molecular beam epitaxy
(MBE). Atomic layer deposition (ALD), originally called atomic layer
epitaxy (ALE), was first reported by Suntola et al. in 1980 for the growth of
zinc sulfide thin films to fabricate electroluminescent flat panel displays.

ALD refers to the method whereby film growth occurs by exposing the
substrate to its starting materials alternately. It should be noted that ALE is
actually a sub-set of ALD, in which the grown film is epitaxial to the
substrate; however, the terms are often used interchangeably. Although both
ALD and CVD use chemical (molecular) precursors, the difference between
the techniques is that the former uses self limiting chemical reactions to
control in a very accurate way the thickness and composition of the film
deposited. In this regard ALD can be considered as taking the best of CVD
(the use of molecular precursors and atmospheric or low pressure) and
MBE (atom-by-atom growth and a high control over film thickness) and
combining them in single method. A selection of materials deposited by
ALD is given in [link].

Compound class

II-VI compounds

II-VI based thin-film
electroluminescent
(TFEL) phosphors

III-V compounds

Semiconductors/dielectric
nitrides

Metallic nitrides
Dielectric oxides
Transparent conductor
oxides

Semiconductor oxides

Superconductor oxides

Fluorides

Examples

ZnS, ZnSe, ZnTe, ZnS;_,Se,, CaS,
SrS, BaS, SrS,;_,Se,, CdS, CdTe,
MntTe, HgTe, Hg,_,Cd,Te,
Cd,_,Mn,Te

ZnS:M (M = Mn, Tb, Tm), CaS:M (M
= Eu, Ce, Tb, Pb), SrS:M (M = Ce,
Tb, Pb, Mn, Cu)

GaAs, AlAs, AIP, InP, GaP, InAs,
Al,Ga,-,As, Ga,Inj_,As, Ga,In,_,P

AIN, GaN, InN, SiN,

TiN, TaN, Ta3Ns, NbN, MoN
AloO3, TiO», ZrOp, HfO,, TajOs,
Nb Os, Y2O3, MgO, CeOsz, SiO»,
La,O3, SrTiO3, BaTiO;

InyO3, In,O3:Sn, InpO3:F, IngO3:Zr,
SnO>, SnO>:Sb, ZnO,

ZnO:Al, GasO3, NiO, CoO,
Y BayCu307_,

CaF), SrF5, ZnF»

Examples of thin film materials deposited by ALD including films
deposited in epitaxial, polycrystalline or amorphous form. Adapted from M.
Ritala and M. Leskel, Nanotechnology, 1999, 10, 19.

How ALD works

The premise behind the ALD process is a simple one. The substrate
(amorphous or crystalline) is exposed to the first gaseous precursor
molecule (elemental vapor or volatile compound of the element) in excess
and the temperature and gas flow is adjusted so that only one monolayer of
the reactant is chemisorbed onto the surface ({link]a). The excess of the
reactant, which is in the gas phase or physisorbed on the surface, is then
purged out of the chamber with an inert gas pulse before exposing the
substrate to the other reactant ({link]b). The second reactant then
chemisorbs and undergoes an exchange reaction with the first reactant on
the substrate surface ((link]c). This results in the formation of a solid
molecular film and a gaseous side product that may then be removed with
an inert gas pulse ({link]d).

(a) Ist precursor pulse

f

Substrate

(d) Purge (c) 2nd precursor pulse
8 Pp Pp

Substrate

Schematic representation of an ALD process.

The deposition may be defined as self-limiting since one, and only one,
monolayer of the reactant species remains on the surface after each
exposure. In this case, one complete cycle results in the deposition of one
monolayer of the compound on the substrate. Repeating this cycle leads to a
controlled layer-by-layer growth. Thus the film thickness is controlled by
the number of precursor cycles rather than the deposition time, as is the
case for a CVD processes. This self-limiting behavior is the fundamental
aspect of ALD and understanding the underlying mechanism is necessary
for the future exploitation of ALD.

One basic condition for a successful ALD process is that the binding energy
of a monolayer chemisorbed on a surface is higher than the binding energy
of subsequent layers on top of the formed layer; the temperature of the
reaction controls this. The temperature must be kept low enough to keep the
monolayer on the surface until the reaction with the second reactant occurs,
but high enough to re-evaporate or break the chemisorption bond. The
control of a monolayer can further be influenced with the input of extra
energy such as UV irradiation or laser beams. The greater the difference
between the bond energy of a monolayer and the bond energies of the
subsequent layers, the better the self-controlling characteristics of the
process.

Basically, the ALD technique depends on the difference between
chemisorption and physisorption. Physisorption involves the weak van der
Waal's forces, whereas chemisorption involves the formation of relatively
strong chemical bonds and requires some activation energy, therefore it may
be slow and not always reversible. Above certain temperatures
chemisorption dominates and it is at this temperature ALD operates best.
Also, chemisorption is the reason that the process is self-controlling and
insensitive to pressure and substrate changes because only one atomic or
molecular layer can adsorb at the same time.

Equipment for the ALD process

Equipment used in the ALD process may be classified in terms of their
working pressure (vacuum, low pressure, atmospheric pressure), method of

pulsing the precursors (moving substrate or valve sources) or according to
the types of sources. Several system types are discussed.

In a typical moving substrate ALD growth system ({link]) the substrate,
located in the recess part of the susceptor, is continuously rotated and cuts
through streams of the gaseous precursors, in this case, trimethylgallium
[TMG, Ga(CH3)3] and arsine (AsH3). These gaseous precursors are
introduced through separate lines and the gases come in contact with the
substrate only when it revolves under the inlet tube. This cycle is repeated
until the required thickness of GaAs is achieved. The exposure time to each
of the gas streams is about 0.3 s.

windows

Substrate in

recess
Rotating part
Quartz tube

reactor

Exhaust

A typical moving substrate ALD growth system used
to grow GaAs films. Adapted from M. A. Tischler
and S. M. Bedair, Appl. Phys. Lett., 1986, 48, 1681.

ALD may be carried out in a vacuum system using an ultra-high vacuum
with a movable substrate holder and gaseous valving. In this manner it may
be also equipped with an in-situ LEED system for the direct observation of
surface atom configurations and other systems such as XPS, UPS, and AES
for surface analysis.

A lateral flow system may also be employed for successful ALE deposition.
This uses an inert gas flow for several functions; it transports the reactants,
it prevents pump oil from entering the reaction zone, it valves the sources
and it purges the deposition site between pulses. Inert gas valving has many
advantages as it can be used at ultra high temperatures where mechanical
valves may fail and it does not corrode as mechanical valves would in the
presence of halides. This method is based on the fact that as the inert gas is
flowing through the feeding tube from the source to the reaction chamber, it
blocks the flow of the sources. Although in this system the front end of the
substrate receives a higher flux density than the down-stream end, a
uniform growth rate occurs as long as the saturation layer of the
monoformation predominates. This lateral flow system effectively utilizes
the saturation mechanism of a monolayer formation obtained in ALE.
Depending on the properties of the precursors used, and on the growth
temperature, various growth systems may be used for ALE.

Requirements for ALD growth

Several parameters must be taken into account in order to assure successful
ALD growth. These include the physical and chemical properties of the
source materials, their pulsing into the reactor, their interaction with the
substrate and each other, and the thermodynamics and volatility of the film
itself.

Source molecules used in ALD can be either elemental or an inorganic,
organic, or organometallic compound. The chemical nature of the precursor
is insignificant as long as it possesses the following properties. It must be a
gas or must volatilize at a reasonable temperature producing sufficient
vapor pressure. The vapor pressure must be high enough to fill the substrate
area so that the monolayer chemisorption can occur in a reasonable length
of time. Note that prolonged exposure to the substrate can cause the

precursor to condense on the surface hindering the growth. Chemical
interaction between the two precursors prior to chemisorption on the
surface is also undesired. This may be overcome by purging the surface
with an inert gas or hydrogen between the pulses. The inert gas not only
separates the reactant pulses but also cleans out the reaction area by
removing excess molecules. Also, the source molecules should not
decompose on the substrate instead of chemisorbing. The decomposition of
the precursor leads to uncontrolled growth of the film and this defeats the
purpose of ALD as it no longer is self-controlled, layer-by-layer growth and
the quality of the film plummets.

In general, temperature remains the most important parameter in the ALD
process. There exists a processing window for ideal growth of monolayers.
The temperature behavior of the rate of growth in monolayer units per cycle
gives a first indication of the limiting mechanisms of an ALD process. If
the temperature falls too low, the reactant may condense or the energy of
activation required for the surface reaction may not be attained. If the
temperature is too high, then the precursor may decompose or the
monolayer may evaporate resulting in poor ALD growth. In the appropriate
temperature window, full monolayer saturation occurs meaning that all
bonding sites are occupied and a growth rate of one lattice unit per cycle is
observed. If the saturation density is below one, several factors may
contribute to this. These include an oversized reactant molecule, surface
reconstruction, or the bond strength of an adsorbed surface atom is higher
when the neighboring sites are unoccupied. Then the lower saturation
density may be thermodynamically favored. If the saturation density is
above one, then the undecomposed precursor molecules form the
monolayer. Generally, ideal growth occurs when the temperature is set
where the saturation density is one.

Advantages of ALD

Atomic layer deposition provides an easy way to produce uniform,
crystalline, high quality thin films. It has primarily been directed towards
epitaxial growth of III-V (13-15) and II-V (12-16) compounds, especially to
layered structures such as superlattices and superalloys. This application is
due to the greatest advantage of this method, it is controllable to an

accuracy of a single atomic layer because of saturated surface reactions.
Not only that, but it produces epitaxial layers that are uniform over large
areas, even on non-planar surfaces, at temperatures lower than those used in
conventional epitaxial growth.

Another advantage to this method that may be most important for future
applications, is the versatility associated with the process. It is possible to
grow different thin films by choosing suitable starting materials among the
thousands of available chemical compounds. Provided that the
thermodynamics are favorable, the adjustment of the reaction conditions is
a relatively easy task because the process is insensitive to small changes in
temperature and pressure due to its relatively large processing window.
There are also no limits in principle to the size and shape of the substrates.

One advantage that is resultant from the self-limiting growth mechanism is
that the final thickness of the film is dependent only upon the number of
deposition cycles and the lattice constant of the material, and can be
reproduced and controlled. The thickness is independent of the partial
pressures of the precursors and growth temperature. Under ideal conditions,
the uniformity and the reproducibility of the films are excellent. ALE also
has the potential to grow very abrupt heterostructures and very thin layers
and these properties are in demand for some applications such as
superlattices and quantum wells.

Bibliography

e D.C. Bradley, Chem. Rev., 1989, 89, 1317.

e M. Ritala and M. Leskel, Nanotechnology, 1999, 10, 19.

e M. Pessa, P. Huttunen, and M. A. Herman, J. Appl. Phys., 1983, 54,
6047.

T. Suntola and J. Antson, Method for producing compound thin films,
U.S. Patent 4,058,430 (1977).

e M.A. Tischler and S. M. Bedair, Appl. Phys. Lett., 1986, 48, 1681.

Chemical Vapor Deposition

Note:This module was developed as part of the Rice University course
CHEM-496: Chemistry of Electronic Materials. This module was prepared
with the assistance of Scott Stokes.

Introduction

Chemical vapor deposition (CVD) is a deposition process where chemical
precursors are transported in the vapor phase to decompose on a heated
substrate to form a film. The films may be epitaxial, polycrystalline or
amorphous depending on the materials and reactor conditions. CVD has
become the major method of film deposition for the semiconductor industry
due to its high throughput, high purity, and low cost of operation. CVD is
also commonly used in optoelectronics applications, optical coatings, and
coatings of wear resistant parts.

CVD has many advantages over physical vapor deposition (PVD) processes
such as molecular beam evaporation and sputtering. Firstly, the pressures
used in CVD allow coating of three dimensional structures with large aspect
ratios. Since evaporation processes are very directional, PVD processes are
typically line of sight depositions that may not give complete coverage due
to shadowing from tall structures. Secondly, high precursor flow rates in
CVD give deposition rates several times higher than PVD. Also, the CVD
reactor is relatively simple and can be scaled to fit several substrates. Ultra-
high vacuum is not needed for CVD and changes or additions of precursors
is an easy task. Furthermore, varying evaporation rates make stoichiometry
hard to control in physical deposition. While for CVD stoichiometry is
more easily controlled by monitoring flow rates of precursors. Other
advantages of CVD include growth of high purity films and the ability to
fabricate abrupt junctions.

There are, however, some disadvantages of CVD that make PVD more
attractive for some applications. High deposition temperatures for some

CVD processes (often greater than 600 °C) are often unsuitable for
structures already fabricated on substrates. Although with some materials,
use of plasma-enhanced CVD or metal-organic precursors may reduce
deposition temperatures. Another disadvantage is that CVD precursors are
often hazardous or toxic and the by-products of these precursors may also
be toxic. Therefore extra steps have to be taken in the handling of the
precursors and in the treatment of the reactor exhaust. Also, many
precursors for CVD, especially the metal-organics, are relatively expensive.
Finally, the CVD process contains a large number of parameters that must
be accurately and reproducibly optimized to produce good films.

Kinetics of CVD

A normal CVD process involves complex flow dynamics since gases are
flowing into the reactor, reacting, and then by-products are exhausted out of
the reactor. The sequence of events during a CVD reaction are shown in
[link] and as follows:

1. Precursor gases input into the chamber by pressurized gas lines.

2. Mass transport of precursors from the main flow region to the substrate
through the boundary layer ([Link]a);

. Adsorption of precursors on the substrate (normally heated) ({link]b).

. Chemical reaction on the surface ([link]c)

. Atoms diffuse on the surface to growth sites.

. Desorption of by-products of the reactions ({link]d).

. Mass transport of by-products to the main flow region ([link]e).

NOD U1 BR W

Main flow of reactant gases

ST

Gaseous by-products

y

Sequence of events during CVD: (a) diffusion of reactants through
boundary layer, (b) adsorption of reactants on substrate, (c)
chemical reaction takes place, (d) desorption of adsorbed species,
and (e) diffusion out of by-products through boundary layer.
Adapted from H. O. Pierson, Handbook of Chemical Vapor
Deposition, Noyes Publications, Park Ridge (1992).

The boundary layer

Gas flow in a CVD reactor is generally laminar, although in some cases
heating of the chamber walls will create convection currents. The complete
problem of gas flow through the system is too complex to be described
here; however, assuming we have laminar flow (often a safe assumption)
the gas velocity at the chamber walls will be zero. Between the wall (zero
velocity) and the bulk gas velocity there is a boundary layer. The boundary
layer thickness increases with lowered gas velocity and the distance from
the tube inlet ({link]). Reactant gases flowing in the bulk must diffuse
through the boundary layer to reach the substrate surface. Often, the

susceptor is tilted to partially compensate for the increasing boundary-layer
thickness and concentration profile.

Reactor cell
SE ae

Susceptor

Development of boundary layer in a
horizontal reactor. Adapted from G. B.
Stringfellow, Organometallic Vapor-Phase
Epitaxy: Theory and Practice, Academic
Press, New York (1994).

Rate limiting steps

During CVD the growth rate of the film is limited by either surface reaction
kinetics, mass transport (diffusion) of precursors to the substrate, or the feed
rate of the precursors.

Surface reaction controls the rate when growth occurs at low temperatures
(where the reaction occurs slowly) and also dominates at low pressures
(where the boundary layer is thin and reactants easily diffuse to the
surface), see [link]. Since reactants easily diffuse through the boundary
layer, the amount of reactant at the surface is independent of reactor
pressure. Therefore, it is the reactions and motions of the precursors
adsorbed on the surface which will determine the overall growth rate of the
film. A sign of surface reaction limited growth would be dependence of the
growth rate on substrate orientation, since the orientation would certainly
not affect the thermodynamics or mass transport of the system.

High gas velocity Low pressure

Low temperature

Rapid diffusion
ete Re ee at onda ayeh
“s

Substrate

Surface reaction limited growth in CVD. Adapted from
H. O. Pierson, Handbook of Chemical Vapor Deposition,
Noyes Publications, Park Ridge (1992).

A deposition limited by mass transport is controlled by the diffusion of
reactants through the boundary layer and diffusion of by-products out of the
boundary layer. Mass transport limits reactions when the temperature and
pressure are high. These conditions increases the thickness of the boundary
layer and make it harder for gases to diffuse through ([link]). In addition,
decomposition of the reactants is typically quicker since the substrate is at a
higher temperature. When mass transport limits the growth, either
increasing the gas velocity or rotating the substrate during growth will

decrease the boundary layer and increase the growth rate.

Low gas velocity High pressure (i.e., atmospheric)

High temperature

Slow diffusion Thick boundary layer

WU ssdédédddssddddddddddsddddddddddsdd

Substrate

Mass transport limited growth in CVD. Adapted from H.

O. Pierson, Handbook of Chemical Vapor Deposition,
Noyes Publications, Park Ridge (1992).

Feed rate limits the deposition when nearly all the reactant is consumed in
the chamber. The feed rate is more important for a hot wall reactor since the
heated walls will decompose a large amount of the precursor. Cold wall
reactors tend to have higher deposition rates since the reactants are not
depleted by the walls.

A plot of growth rate versus temperature, known as an Arrhenius plot, can
be used to determine the rate limiting step of a reaction ([link]). Mass
transport limits reactions at high temperatures such that growth rate
increases with partial pressures of reactants, but is constant with
temperature. Surface reaction kinetics dominates at low temperatures where
the growth rate increases with temperature, but is constant with pressures of
reactants. Feed rate limited reactions are independent of temperature, since
it is the rate of gas delivery that is limiting the reaction. The Arrhenius plot
will show where the transition between the mass transport limited and the
surface kinetics limited growth occurs in the temperature regime.

Gas-phase-transport
: or feed-rate limited :

Surface-reaction limited

Deposition rate
(arb units)

Precursor
depletion

1/T

Dependence of CVD deposition rate on temperature.
Adapted from J. G. Eden, in Thin Film Processes IT, Eds.
J. L. Vossen and W. Kern, Academic Press, New York
(1991).

Increases in reactant concentrations will to a point increase the deposition
rate. However, at very high reactant concentrations, gas phase nucleation
will occur and the growth rate will drop ([link]). Slow deposition in a CVD
reactor can often be attributed to either gas phase nucleation, precursor
depletion due to hot walls, thick boundary layer formation, low
temperature, or low precursor vapor pressure.

Deposition rate
(arb units)

Surface ‘ Nucleation in gas phase

reaction
limited

Reactant concentration

Demonstration of deposition rate on reactant
concentration for CVD deposition. Adapted from J. G.
Eden, in Thin Film Processes II, Eds. J. L. Vossen and

W. Kern, Academic Press, New York (1991).

CVD systems

Precursor delivery

Flow of reactants into the reactor must be closely monitored to control
stoichiometry and growth rate. Precursor delivery is very important since in
many cases the flow rate can limit the deposition. For low vapor pressure
solids, a carrier gas is passed over or through a bed of the heated solid to

transport the vapor into the reactor. Gas flow lines are usually heated to
reduce condensation of the vapor in the flow lines. In the case of gas
precursors, mass flowmeters easily gauge and control the flow rates. Liquid
precursors are normally heated in a bubbler to achieve a desired vapor
pressure ([link]).

Carrier gas ——=—

Carrier gas and
a ell
_—_—<— reactant vapor

Liquid or
molten precursor

Schematic representation of a bubbler for
liquid precursors.

An inert gas such as hydrogen is bubbled through the liquid and by
calculating the vapor pressure of the reactant and monitoring the flow rate
of the hydrogen, the flow rate of the precursor is controlled by using [link],
where Qyo is the flow rate of the metal-organic precursor, Qyp is the flow
rate of hydrogen through the bubbler, Pyyg is the vapor pressure of the
metal-organic at the bubbler temperature, and Pp is the pressure of the
bubbler.

Equation:

Pao

Q0= Was —
" - Pg - Péo

Another method of introducing liquid precursors involves flash
vaporization where the liquid is passed into a flask heated above the boiling

point of the liquid. The gas vapor is then passed through heated lines to the
CVD chamber. Often, a carrier gas is added to provide a fixed flow rate into
the reactor. This method of precursor introduction is useful when the
precursor will decompose if heated over time. A similar technique called
spray pyrolysis introduces the precursors in the form of aerosol droplets.
The droplets evaporate in the chamber from the heated gas above the
substrate or heated chamber walls ([link]). Then the reaction proceeds as a
normal CVD process.

Solution of
precursor

o.oo. ee eee

———_ Atomizer siresiststststsss: :

Precursor
evaporation

carrier gas
Substrate

Heater

Schematic representation of a typical aerosol delivery
system for CVD precursors. Adapted from T. T. Kodas
and M. J. Hamton-Smith, The Chemistry of Metal CVD,
VCH, New York (1994).

Thermal CVD reactors

In thermal CVD temperatures as high as 2000 °C may be needed to
thermally decompose the precursors. Heating is normally accomplished by
use of resistive heating, radio frequency (rf) induction heating, or radiant
heating. There are two basic types of reactors for thermal CVD: the hot wall
reactor and the cold wall reactor.

A hot wall reactor is an isothermal furnace into which the substrates are
placed. Hot wall reactors are typically very large and depositions are done
on several substrates at once. Since the whole chamber is heated, precise
temperature control can be achieved with correct furnace design. A
disadvantage of the hot wall configuration is that deposition occurs on the
walls of the chamber as well as on the substrate. As a consequence, hot wall
reactors must be frequently cleaned to reduce flaking of particles from the
walls which may contaminate the substrates. Furthermore, reactions in the
heated gas and at the walls deplete the reactants and can result in feed rate
limited growth. [link] shows a typical low pressure hot wall CVD reactor.

Pressure sensor

Resistance heater (3-zone)

—» Exhaust

Tray and wafers

Gas inlet

Schematic of a typical low pressure hot wall CVD
reactor used in coating silicon substrates. Adapted from
H. O. Pierson, Handbook of Chemical Vapor Deposition,
Noyes Publications, Park Ridge (1992).

In a cold wall reactor only the substrate is heated, usually by induction or
radiant heating. Since most CVD reactions are endothermic, deposition is
preferentially on the area of highest temperature. As a result, deposition is
only on the substrate and the cooler reactor walls stay clean. Cold wall
CVD has two main advantages over the hot wall configuration. First,

particulate contamination is reduced since there are no deposits formed on
the walls of the reactor. Second, since decomposition only occurs on the
substrate there is no depletion of source gases due to reaction on the walls.
However, hot wall reactors tend to have higher throughput since the designs
more easily accommodate multiple wafer configurations.

The final issue in design of a thermal CVD reactor is the operating pressure.
The pressure of the reactor has a large effect on the rate limiting step of the
deposition. Atmospheric pressure reactors have a large boundary layer
({link]) and non-uniform diffusion of reactants through the boundary layer
often results in non-uniform film compositions across the wafer.
Conversely, low pressure reactors have a nearly non-existent boundary later
and reactants easily diffuse to the substrate ((link]). However, the difficulty
in maintaining a uniform temperature profile across the wafer can result in
thickness non-uniformities since the deposition rate in low pressure reactors
is strongly temperature dependent. Careful studies of the flow dynamics and
temperature profiles of CVD reactors are always carried out in order to
achieve uniform material depositions.

Plasma-enhanced CVD

Plasmas are generated for a variety of thin film processes including
sputtering, etching, ashing, and plasma-enhanced CVD. Plasma-enhanced
CVD (PECVD), sometimes called plasma-assisted (PAC VD), has the
advantage that plasma activated reactions occur at much lower temperatures
compared to those in thermal CVD. For example, the thermal CVD of
silicon nitride occurs between 700 - 900 °C, the equivalent PECVD process
is accomplished between 250 - 350 °C.

A plasma is a partially ionized gas consisting of electrons and ions. Typical
ionization fractions of 10° to 10°! are encountered in process reactors.
Plasmas are electrically conductive with the primary charge carriers being
the electrons. The light mass of the electron allows it to respond much more
quickly to changes in the field than the heavier ions. Most plasmas used for
PECVD are generated using a rf electric field. In the high frequency electric
field, the light electrons are quickly accelerated by the field but do not

increase the temperature of the plasma because of their low mass. The
heavy ions cannot respond to the quick changes in direction and therefore
their temperature stays low. Electron energies in the plasma have a
Maxwellian distribution in the 0.1 — 20 eV range. These energies are
sufficiently high to excite molecules or break chemical bonds in collisions
between electrons and gas molecules. The high energy electrons
inelastically collide with gas molecules resulting in excitation or ionization.
The reactive species generated by the collisions do not have the energy
barriers to reactions that the parent precursors do. Therefore, the reactive
ions are able to form films at temperatures much lower than those required
for thermal CVD.

The general reaction sequence for PECVD is shown in [link]. In addition to
the processes that occur in thermal CVD, reactive species resulting from
electron dissociation of parent molecules also diffuse to the surface. The
reactive species have lower activation energies for chemical reactions and
usually have higher sticking coefficients to the substrate. Therefore, the
PECVD reaction is dominated by the reactive species on the surface and not
any of the the parent precursor molecules that also diffuse to the surface.

Plasma: *: Neutral species.‘ . ‘Ionic species +-1+l+leleleletet

Boundary layer ©" 2+ Si Sate 2 ee Ue oo eo 2 et
Diffusion Acceleration Desorption,
chemical sputtering

Sheath
Ion bombardment,

Migration adsorption, dissociation Reaction

ddd lll

Reaction sequence in PECVD. Adapted from M.
Konuma, Film Deposition by Plasma Techniques,
Springer-Verlag, New York (1992).

A basic PECVD reactor is shown in [link]. The wafer chuck acts as the
lower electrode and is normally placed at ground potential. Gases are either
introduced radially at the edges of the reactor and pumped out from the
center, or gases can be introduced from the center and pumped at the edges
as shown in [link]. The magnetic drive allows rotation of the wafers during
processing to randomize substrate position. Some newer reactors introduce
the gases through holes drilled in the upper electrode. This method of gas
introduction gives a more uniform plasma distribution.

Input from shielded rf power

|

Electrode

Silicon wafers

Rotating

shaft To vacuum

pump and exhaust

To vacuum
pump and exhaust

| Magnetic drive
Gases

Schematic representation of a radial flow PECVD
reactor. Adapted from H. O. Pierson, Handbook of
Chemical Vapor Deposition, Noyes Publications, Park
Ridge (1992).

Plasma CVD has numerous advantages over thermal CVD. Obviously the
reduced deposition temperature is a bonus for the semiconductor industry
which must worry about dopant diffusion and metal interconnects melting
at the temperatures required for thermal CVD. Also, the low pressures
(between 0.1 - 10 Torr) required for sustaining a plasma result in surface
kinetics controlling the reaction and therefore greater film uniformity. A
disadvantage of plasma CVD is that it is often difficult to control
stoichiometry due to variations in bond strengths of various precursors. For
example, PECVD films of silicon nitride tend to be silicon rich because of
the relative bond strength of N> relative to the Si-H bond. Additionally,
some films may be easily damaged by ion bombardment from the plasma.

Photochemical CVD

Photochemical CVD uses the energy of photons to initiate the chemical
reactions. Photodissociation of the chemical precursor involves the
absorption of one or more photons resulting in the breaking of a chemical
bond. The most common precursors for photo-assisted deposition are the
hydrides, carbonyls, and the alkyls. The dissociation of dimethylzinc by
[link], a photon creates a zinc radical and a methyl radical (‘CH3) that will
react with hydrogen in the reactor to produce methane.

Equation:

Zn(CH;), + hv (64eV) + H, > Zn + 2CH,

Like several metal-organics, dimethylzinc is dissociated by the absorption
of only one UV photon. However, some precursors require absorption of
more than one photon to completely dissociate. There are two basic
configurations for photochemical CVD. The first method uses a laser
primarily as a localized heat source. The second method uses high energy
photons to decompose the reactants on or near the growth surface.

In thermal laser CVD, sometimes referred to as laser pyrolysis, the laser is
used to heat a substrate that absorbs the laser photons. Laser heating of
substrates is a very localized process and deposition occurs selectively on
the illuminated portions of the substrates. Except for the method of heating,
laser CVD is identical to thermal CVD. The laser CVD method has the
potential to be used for direct writing of features with relatively high
resolution. The lateral extent of film growth when the substrate is
illuminated by a laser is determined not only by the spot size of the laser,
but by the thermal conductivity of the substrate. A variation of laser
pyrolysis uses a laser to heat the gas molecules such that they are
fragmented by thermal processes.

Photochemical effects can be induced by a laser if the precursor molecules
absorb at the laser wavelength. UV photons have sufficient energy to break
the bonds in the precursor chemicals. Laser-assisted CVD (LACVD) uses a
laser, usually an eximer laser, to provide the high energy photons needed to
break the bonds in the precursor molecules. [link] shows two geometries for
LACVD. For the perpendicular illumination the photochemical effects
generally occur in the adsorbed adlayer on the substrate. Perpendicular
irradiation is often done using a UV lamp instead of a laser so that
unwanted substrate heating is not produced by the light source. The parallel
illumination configuration has the benefit that reaction by-products are
produced further from the growth surface and have less chance of being
incorporated into the growing film. The main benefit of LACVD is that
nearly no heat is required for deposition of high quality films.

(a) Gas

Window Reactor
Laser
Substrate
Vacuum system
(b) Gas
Window Reactor
Laser
Window

Vacuum system

Parallel (a) and perpendicular (b) irradiation
in laser CVD. Adapted from J. G. Eden, in
Thin Film Processes II, Eds. J. L. Vossen
and W. Kern, Academic Press, New York
(1991).

An application of laser photolysis is photonucleation. Photonucleation is the
process by which a chemisorbed adlayer of metal precursors is photolyzed

by the laser to create a nucleation site for further growth. Photonucleation is
useful in promoting growth on substrates that have small sticking
coefficients for gas phase metal atoms. By beginning the nucleation process
with photonucleation the natural barrier to surface nucleation on the
substrate is overcome.

Bibliography

J. G. Eden, in Thin Film Processes IT, Eds. J. L. Vossen and W. Kern,
Academic Press, New York (1991).

T. T. Kodas and M. J. Hamton-Smith, The Chemistry of Metal CVD,
VCH, New York (1994).

M. Konuma, Film Deposition by Plasma Techniques, Springer-Verlag,
New York (1992).

H. O. Pierson, Handbook of Chemical Vapor Deposition, Noyes
Publications, Park Ridge (1992).

R. Reif and W. Kern, in Thin Film Processes II, Eds. J. L. Vossen and
W. Kern, Academic Press, New York (1991).

G. B. Stringfellow, Organometallic Vapor-Phase Epitaxy: Theory and
Practice, Academic Press, New York (1994).

Liquid Phase Deposition

Introduction

Silicon dioxide (silica, SiOz) has been the most researched chemical
compound apart from water. Silica has been used throughout history, for
example, flint, which when sharpened formed one of humanities first tools.
Crystalline silica, or sand, was melted into glass as early as 5000 B.C.,
birthing a technology that has gained sophistication in modern times.
Silicon is the second most plentiful element in the Earth’s crust, the most
plentiful being oxygen. It is thus surprising that it was not until 1800 that
silica was named a compound by Sir Humphry Davy. He, however, failed to
isolate its components via electrolysis, and it is Jons Jacob Berzelius who is
thus credited with discovering silica in 1824. He heated potassium
fluorosilicate with potassium metal and, after purifying the product of this
reaction with water, produced amorphous silica powder.

The most common forms of silica employed in industry include a-quartz,
vitreous silica, silica gel, fumed silica and diatomaceous earth. Synthetic
quartz is hydrothermally grown from a seed crystal, with aqueous NaOH
and vitreous SiO», at 400 °C and 1.7 kbar. Because it is a piezoelectric
material, it is used in crystal oscillators, transducers, pickups and filters for
frequency control and modulation. Vitreous silica is super cooled liquid
silica used in laboratory glassware, protective tubing sheaths and vapor
grown films. Silica gel is formed from the reaction of aqueous sodium
silicate with acid, after which it is washed and dehydrated. Silica gel is an
exceptionally porous material with numerous applications including use as
a dessicant, chromatographic support, catalyst substrate and insulator.
Pyrogenic or fumed silica is produced by the high temperature hydrolysis,
in an oxyhydrogen flame, of SiCly. Its applications include use as a
thickening agent and reinforcing filler in polymers. Diatomaceous earth, the
ecto-skeletons of tiny unicellular marine algae called diatoms, is mined
from vast deposits in Europe and North America. Its primary use is in
filtration. Additional applications include use as an abrasive, insulator, filler
and a lightweight aggregate.

Methods of colloidal growth and thin film deposition of amorphous silica
have been investigated since 1925. The two most common and well-
investigated methods of forming SiO, in a sol or as a film or coating are
condensation of alkoxysilanes (known as the Stober method) and hydrolysis
of metal alkoxides (the Iler or dense silica [DS] process).

Liquid phase deposition (LPD)

LPD is a method for the “non-electrochemical production of polycrystalline
ceramic films at low temperatures.” LPD, along with other aqueous solution
methods [chemical bath deposition (CBD), successive ion layer adsorption
and reaction (SILAR) and electroless deposition (ED) with catalyst] has
developed as a potential substitute for vapor-phase and chemical-precursor
systems. Aqueous solution methods are not dependent on vacuum systems
or glove boxes, and the use of easily acquired reagents reduces reliance on
expensive or sensitive organometallic precursors. Thus, LPD holds potential
for reduced production costs and environmental impact. Films may be
deposited on substrates that might not be chemically or mechanically stable
at higher temperatures. In addition, the use of liquid as a deposition medium
allows coating of non-planar substrates, expanding the range of substrates
that are capable of being coated. Aqueous deposition techniques have not
reached the level of maturation that vapor-phase techniques have in respect
to a high level of control over composition, microstructure and growth rates
of the resulting films, but their prospect makes them attractive for research.

LPD generally refers to the formation of oxide thin films, the most common
being SiO>, from an aqueous solution of a metal-fluoro complex [MF,]™”,
which is slowly hydrolyzed using water, boric acid or aluminum metal.
Addition of water drives precipitation of the oxide. Boric acid and
aluminum work as fluoride scavengers, rapidly weakening the fluoro
complex and precipitating the oxide. These reactants are added either drop
wise or outright, both methods allowing for high control of the hydrolysis
reaction and of the solution’s supersaturation. Film formation is
accomplished from highly acidic solutions, in contrast to the basic or
weakly acidic solutions used in chemical bath deposition.

A generic description of the LPD reaction is shown in [link], where m is the
charge on the metal cation. If the concentration of water is increased or the
concentration of hydrofluoric acid (HF) is decreased, the equilibrium will
be shifted toward the oxide. Use of boric acid or aluminum metal will
accomplish the latter, see [link] and [link]. The most popular of these
methods for accomplishing oxide formation has been through the addition
of boric acid.

Equation:

Equation:

H,BO,+4HF=—= BF, +H,0*+2H,O
Equation:

Al+6HF === H,AIF, + 1.5 H,

The first patent using liquid phase deposition (LPD) of silicon dioxide via
fluorosilicic acid solutions (H»SiF,) was granted to the Radio Corporation
of America (RCA) in 1950. RCA used LPD as a method for coating anti-
reflective films on glass, but the patent promised further applications. Since
this initial patent there have been many further patents and papers utilizing
this method, in variable forms, to coat substrates, usually silicon, with
silicon dioxide. The impetus behind this work is to create an alternative to
the growth of insulator coatings by thermal oxidation or chemical vapor
deposition (CVD) for planar silicon chip technology. Thermal oxidation and
CVD are performed at elevated temperatures, requiring a higher output of
energy and more complicated instrumentation than that of LPD. The most
simple and elegant of the LPD methods uses only water to catalyze silica
thin film growth on silicon from a solution of fluorosilicic acid
supersaturated with silicon dioxide, [link].

Equation:

H,SiF,+2H,O === SiO, +6HF

The amount of water reacted with the supersaturated fluorosilicic acid
solution controls both the growth rate and incorporation of fluorine into the
resulting silica matrix. Both growth rate and fluorine content increase with
increased addition of water. Ultimately this “dilution” affects the optical
properties of the resulting silica film; an increased amount of fluorine
decreases its dielectric constant (and thus its refractive index).

To ensure a uniform film growth with LPD, the preparation of the surface to
be coated is of utmost importance. Suitable treatments may involve the
formation of surface hydroxides, the pre-deposition or self-assembly of an
appropriate seed layer. The most efficient coverage is seen with silicon
surfaces functionalized with hydroxy (-OH) groups prior to immersion in
the growth solution. This can be achieved through appropriate etching of
the silicon surface. It is proposed that the silanol (Si-OH) groups act to seed
the growth of the silica film through condensation reactions with the silicic
acid formed in the growth solution.

Lee and co-workers and Homma separately propose that intermediate,
hydrolyzed species, SiF,(OH)4_, (n < 4), are formed by the reaction shown
in [link]. According to Lee, these species then react with the substrate
surface to form a film. Homma proposes that fluorine-containing siloxanes
are subsequently formed, which adsorb onto the surface where
condensation and bonding occurs between the oligomers and surface
hydroxyl groups. The former mechanism implies a molecular growth
mechanism, whereas the latter implies homogeneous nucleation with
subsequent deposition.

Equation:

H,SiF, + (4-n)H,0 === SiF,(OH),,, + (6-n) HF
In concentrated fluorosilicic acid solutions silica can be dissolved to well

beyond its solubility, forming fluorosilicon complexes such as [SiFg.SiF4]*,
[link]. The bridged fluorosilicon complex has electron deficient silicon

because of the high electronegativity of the bonded fluorines, creating weak
Si-F bonds. These bonds are then prone to nucleophilic attack by water. The
fluorine ion (F-) combines with the proton in this reaction to form
hydrofluoric acid (HF). The product of this reaction can then react further
with water to yield [SiF,(OH),]*, SiF, and HF. The high acidity of the
solution then allows protons to react with [SiF,(OH),|*" to form
tetrafluorosilicate (SiF,4) and water, [link]. Hydrolysis of the SiF, will then
yield the hexafluorosilicate anion, protons and silicic acid, [link].
Equation:

5 H,SiF, + SiO, > 3 [SiF,SiF,|> + 2H,O + 6 H*
Equation:

[SiF,(OH),|? + 2H* — SiF,+ 2H,O
Equation:

3 SiF,+4H,O — 2 SiF,* + Si(OH), + 4 H*

Silicic acid is adsorbed onto the surface of the substrate that has been
introduced into the growth solution. Molecular growth of silica on the
substrate surface is initialized in an acid catalyzed dehydration between the
silicic acid and the silanol groups on the substrate surface. Si-O-Si bonds
are formed, resulting in an initial silica coating of the surface. Following
reactions between the initial silica coating and the monosilicic acid in
solution result in further silica deposition and growth. Because of the
presence of HF in the solution, the surface and growing silica matrix is
subject to attack according to the reaction in [link]. This explains the
incorporation of a quantity of fluorine into the silica film. Additionally, it
reveals that a certain amount of silica etching occurs along with growth.
Because of the prevalence of the silicic acid in the solution, however,
deposition is predominant.

Equation:

Si-OH + HF — Si-F+H,O

This proposed mechanism, which is more in depth than those proposed by
Lee and Homma, elucidates what is experimentally proven. The deposition
rate of the silica increases with addition of H»O because the nucleophilic
attack of the fluorosilicon complex is then augmented, increasing the
concentration of silicic acid in the growth solution. The H2O addition
increases the reaction rate and thus the concentration of HF in the growth
solution, resulting in greater incorporation of fluorine into the silica matrix
because of HF attack of the deposited film. Additionally, Yeh’s mechanism
supports a molecular growth model, i.e., heterogeneous growth, which
represents a consensus of the body of research performed thus far.

In a solution with dissolved ceramic precursors, nucleation and growth will
occur either in solution (homogenous nucleation) or on the surfaces of
introduced solid phases (heterogeneous nucleation). Successful film
formation relies on the promotion of heterogeneous nucleation. Solubility
generally depends on the solution pH and the concentration of the species in
solution. As the solution crosses over from a solvated state to a state of
supersaturation, film formation can occur. It is vital to assure that the state
of supersaturation is one that promotes film growth and not homogeneous
nucleation and precipitation. This concept is illustrated in [link].

precipitation

log [M]

saturation limit

soluble

pH

Idealized solubility diagram for film forming
species in water. Adapted from B. C. Bunker, P. C.
Rieke, B. J. Tarasevich, A. A. Campbell, G. E.
Fryxall, G. L. Graff, L. Song, J. Liu, J. W. Virden,
and G. L. McVay, Science, 1994, 264, 48.

Silica can be dissolved in fluorosilicic acid to well above its solubility in
water, which is approximately 220 ppm (mg/L). Depending on the
concentration of the fluorosilicic acid solution, it can contain up to 20%
more silica than is implied by the formula H»SiF,. After saturation of the
solution with SiO», the solvated species is a mixture of fluorosilicates,
which reacts as explained earlier. It must be emphasized that addition of
water in this reaction is not simply dilution, but is the addition of a reactant,
which places the solution in a metastable state (the blue area in [link]) in
preparation for the introduction of a suitable surface to seed the growth of
silica.

Another important factor in solution growth methods is interfacial energy.
When a substrate with lower interfacial energy than that of a growing
homogeneous nucleus is introduced into a growth solution, heterogeneous
growth is favored. Thus, a seeded growth mechanism by definition
introduces a substrate of lower interfacial energy into a supersaturated
solution, facilitating heterogeneous growth. Lower interfacial energies can
be a product of surface modification, as well as a property of the materials’
natural state.

Comparing LPD to sol-gel

An alternative method to LPD for forming silica thin films is the sol-gel
method. A sol is a colloidal dispersion of particles in a liquid. A gel is a
material that contains a continuous solid matrix enclosing a continuous
liquid phase. The liquid inhibits the solid from collapsing and the solid
impedes release of the liquid. A formal definition of sol-gel processing is
the “growth of colloidal particles and their linking together to form a gel.”
This method describes both the hydrolysis and condensation of silicon
alkoxides and the hydrolysis and condensation of aqueous silicates (the DS
process).

In the hydrolysis of silicon alkoxides, an alkoxide group is replaced with a
hydroxyl group, [link]. Further condensation reactions between alkoxyl

groups or hydroxyl groups produce siloxane bonds, see [link] and [link].
Equation:

=Si-OR+H,O =— =Si-OH + ROH
Equation:
=Si-OH + RO-Siz ==—* =Si-O-Si= + ROH

Equation:

=Si-OH+HO-Sie =— =Si-O-Si= +H,O

Tetramethoxysilane [Si(QMe)4, TMOS] and tetraethylorthoxysilane
[Si(OEt)4, TEOS] are the most commonly used precursors in silica sol-gel
processing. The alkoxides are hydrolyzed in their parent alcohols, with a
mineral acid or base catalyst, producing silicate gels that can be deposited
as coatings. The Stober method, which utilizes this chemistry, relies on
homogeneous nucleation to produce monodisperse sols.

Iler’s DS method of silica film formation was originally patented as a
pigment coating to increase dispersibility of titania particles for use in the
paint industry. The DS method is based on the aqueous chemistry of silica
and takes advantage of the species present in solution at varying pH. Below
pH 7 three-dimensional gel networks are formed. Above pH 7 silica
surfaces are quite negatively charged ([link]), so that particle growth occurs
without aggregation. The isoelectric point of silica is pH 2. Reactions above
and below pH 2 are thought to occur through bimolecular nucleophilic
condensation mechanisms. Above pH 2 an anionic species attacks a neutral
species ([link]) and below pH 2 condensation involves a protonated silanol
({link]). The DS process has been utilized extensively in sol-gel coating
technology and as a growth method for monodisperse and polydisperse sols.
Equation:

Si(OH), (aq) —> Si(OH),O° + H*
Equation:

SiO’ + =Si-OH —> =Si-O-Si= +OH-
Equation:

=SiOH,* + HO-Si= — =Si-O-Si= + H*

Bibliography

e B.C. Bunker, P. C. Rieke, B. J. Tarasevich, A. A. Campbell, G. E.
Fryxall, G. L. Graff, L. Song, J. Liu, J. W. Virden, and G. L. McVay,

Science, 1994, 264, 48.

P.-H. Chang, C.-T. Huang, and J.-S. Shie, J. Electrochem. Soc., 1997,
144, 1144.

J.-S. Chou and S.-C. Lee, J. Electrochem. Soc., 1994, 141, 3214.

T. Homma, T. Katoh, Y. Yamada, and Y. Murao, J. Electrochem. Soc.,
1993, 140, 2410.

R. K. Iler, The Chemistry of Silica Solubility, Polymerization, Colloid
and Surface Properties, and Biochemistry, John Wiley & Sons (1979).
H. R. Jafry, E. A. Whitsitt, and A. R. Barron, J. Mater. Sci., 2007, 42,
7381.

T. Niesen and M. R. De Guire, J. Electroceramics, 2001, 6, 169.

N. Ozawa, Y. Kumazawa, and T. Yao, Thin Solid Films, 2002, 418,
102.

W. Stober, A. Fink, and E. Bohn, J. Colloid Interface Sci., 1968, 26,
62.

D. Whitehouse, Glass of the Roman Empire, Corning (1988).

E. A. Whitsitt and A. R. Barron, Nano Lett., 2003, 3, 775.

E. A. Whitsitt and A. R. Barron, Chem. Commun., 2003, 1042.

E. A. Whitsitt and A. R. Barron, J. Colloid Interface Sci., 2005, 287,
318.

C.-F. Yeh, C.-L. Chen, and G.-H. Lin, J. Electrochem. Soc., 1994, 141,
3177.

Selecting a Molecular Precursor for Chemical Vapor Deposition

Introduction

The proven utility of chemical vapor deposition (CVD) in a wide range of
electronic materials systems (semiconductors, conductors, and insulators)
has driven research efforts to investigate the potential for thin film growth
of other materials, including: high temperature superconducting metal
oxides, piezoelectric material, etc. Moreover, CVD potentially is well suited
for the preparation of thin films on a wide range of substrates, including
those of nonplanar geometries. CVD offers the advantages of mild process
conditions (i.e., low temperatures), control over microstructure and
composition, high deposition rates, and possible large scale processing. As
with any CVD process, however, the critical factor in the deposition process
has been the selection of precursors with suitable transport properties.

Factors in selecting a CVD precursor molecule

The following properties are among those that must be considered when
selecting suitable candidates for a CVD precursor:

1. The precursor should be either a liquid or a solid, with sufficient vapor
pressure and mass transport at the desired temperature, preferably
below 200 °C. Liquids are preferred over solids, due to the difficulty
of maintaining a constant flux of source vapors over a non-equilibrium
percolation (solid) process. Such non-bubbling processes are a
function of surface area, a non-constant variable with respect both to
time and particle size. The upper temperature limit is not dictated by
chemical factors; rather, it is a limitation imposed by the stability of
the mass flow controllers and pneumatic valves utilized in commercial
deposition equipment. It must be stressed that while the achievement
of an optimum vapor pressure for efficient utilization as an industrially
practicable source providing high film growth rates (>10 Torr at 25 °C)
is a worthy goal, the usable pressure regimes are those in which
evaluation can be carried out on compounds which exhibit vapor
pressures exceeding 1 Torr at 100 °C.

2. The precursor must be chemically and thermally stable in the region
bordered by the evaporation and transport temperatures, even after
prolonged use. Early workers were plagued by irreproducible film
growth results caused by premature decomposition of source
compounds in the bubbler, in transfer lines, and, basically everywhere
except on the substrate. Such experiences are to be avoided!

3. By its very nature, CVD demands a decomposable precursor. This
generally is accomplished thermally; however, the plasma-enhanced
growth regime has seen much improvement. In addition, photolytic
processes have tremendous potential. Nevertheless, the precursor must
be thermally robust until deposition conditions are employed.

4. The precursor should be relatively easy to synthesize, ensuring
sufficient availability of material for testing and fabrication. It also is
important that the synthesis of the compound be reproducible. It
should be simple to prepare and purify to a relatively high level of
purity. It should be non-toxic and environmentally friendly (i.e., as low
a toxicity as can be attained, given the fundamental toxicity of
particular elements such as mercury, thallium, barium, etc.). It should
be routine to reproduce and scale-up the preparation for further
developmental studies. It should utilize readily available starting
reagents, and proceed by a minimum number of chemical
transformations in order to minimize the cost.

5. Due to handling considerations, the source should be oxidatively,
hydrolytically, thermally and photochemically stable under normal
storage conditions, in addition the precusor should resist
oligomerization (in the solid, liquid, or gaseous states). It is worth
noting that practitioners of metal organic CVD (MOCVD), especially
for 13-15 materials have of necessity become expert in the handling of
very toxic, highly air sensitive materials.

Historically, researchers were limited in their choices of precursors to those
that were readily known and commercially available. It must be emphasized
that none of these previously known compounds had been designed
specifically to serve as vapor phase transport molecules for the associated
element. Thus, the scope was often limited to what was commercially
available. However, as new compounds have now been made with the

specific goal of providing ideal CVD precursors the choice to academia and
industry has increased.

Bibliography

e G. B. Stringfellow, Organometallic Vapor Phase Epitaxy: Theory and
Practice, Academic Press, New York (1989).

Determination of Sublimation Enthalpy and Vapor Pressure for Inorganic
and Metal-Organic Compounds by Thermogravimetric Analysis

Introduction

Metal compounds and complexes are invaluable precursors for the chemical
vapor deposition (CVD) of metal and non-metal thin films. In general, the
precursor compounds are chosen on the basis of their relative volatility and
their ability to decompose to the desired material under a suitable
temperature regime. Unfortunately, many readily obtainable (commercially
available) compounds are not of sufficient volatility to make them suitable
for CVD applications. Thus, a prediction of the volatility of a metal-organic
compounds as a function of its ligand identity and molecular structure
would be desirable in order to determine the suitability of such compounds
as CVD precursors. Equally important would be a method to determine the
vapor pressure of a potential CVD precursor as well as its optimum
temperature of sublimation.

It has been observed that for organic compounds it was determined that a
rough proportionality exists between a compound’s melting point and
sublimation enthalpy; however, significant deviation is observed for
inorganic compounds.

Enthalpies of sublimation for metal-organic compounds have been
previously determined through a variety of methods, most commonly from
vapor pressure Measurements using complex experimental systems such as
Knudsen effusion, temperature drop microcalorimetry and, more recently,
differential scanning calorimetry (DSC). However, the measured values are
highly dependent on the experimental procedure utilized. For example, the
reported sublimation enthalpy of Al(acac)3 ([link]a, where M = Al, n = 3)
varies from 47.3 to 126 kJ/mol.

Structure of a typical metal B-diketonate

complex. (a) acetylacetonate (acac); (b)

trifluoro acetylacetonate (tfac), and (c)
hexafluoroacetylacetonate (hfac).

Thermogravimetric analysis offers a simple and reproducible method for
the determination of the vapor pressure of a potential CVD precursor as
well as its enthalpy of sublimation.

Determination of sublimation enthalpy

The enthalpy of sublimation is a quantitative measure of the volatility of a
particular solid. This information is useful when considering the feasibility
of a particular precursor for CVD applications. An ideal sublimation
process involves no compound decomposition and only results in a solid-
gas phase change, i.e., [link].

Equation:

IM(L),, I sotia) a IM(L), | vapor)
Since phase changes are thermodynamic processes following zero-order

kinetics, the evaporation rate or rate of mass loss by sublimation (m,,p), at a
constant temperature (T), is constant at a given temperature, [link].

Therefore, the m,,, values may be directly determined from the linear mass
loss of the TGA data in isothermal regions.
Equation:

m,,» = _Almass]
At

The thermogravimetric and differential thermal analysis of the compound
under study is performed to determine the temperature of sublimation and
thermal events such as melting. [link] shows a typical TG/DTA plot for a
gallium chalcogenide cubane compound ({link]).

100
= 80
=
=
r a
5 50 >
< =
_—
&
—_*
)
0 105 210 315 420

Temperature (°C)

A typical thermogravimetric/differential
thermal analysis (TG/DTA) analysis of
[(EtMe,C)GaSe],, whose structure is shown
in [link]. Adapted from E. G. Gillan, S. G.
Bott, and A. R. Barron, Chem. Mater., 1997,
9, 3, 796.

Structure of
gallium
chalcogenide
cubane
compound,
where E = S,
Se, and R =
CMes,
CMe>Ft,
CEt Me,
CEts.

Data collection

In a typical experiment 5 - 10 mg of sample is used with a heating rate of
ca. 5 °C/min up to under either a 200-300 mL/min inert (N> or Ar) gas flow
or a dynamic vacuum (ca. 0.2 Torr if using a typical vacuum pump). The
argon flow rate was set to 90.0 mL/min and was carefully monitored to
ensure a steady flow rate during runs and an identical flow rate from one set
of data to the next.

Once the temperature range is defined, the TGA is run with a
preprogrammed temperature profile ({link]). It has been found that
sufficient data can be obtained if each isothermal mass loss is monitored
over a period (between 7 and 10 minutes is found to be sufficient) before

moving to the next temperature plateau. In all cases it is important to
confirm that the mass loss at a given temperature is linear. If it is not, this
can be due to either (a) temperature stabilization had not occurred and so
longer times should be spent at each isotherm, or (b) decomposition is
occurring along with sublimation, and lower temperature ranges must be
used. The slope of each mass drop is measured and used to calculate
sublimation enthalpies as discussed below.

Temperature (°C)

Isotherm

0 10 20 30
Time (min.)

A typical temperature profile for
determination of isothermal mass loss rate.

As an illustrative example, [link] displays the data for the mass loss of
Cr(acac)3 ([link]a, where M = Cr, n = 3) at three isothermal regions under a
constant argon flow. Each isothermal data set should exhibit a linear
relation. As expected for an endothermal phase change, the linear slope,
equal to Mgyp, increases with increasing temperature.

14.2

B14.
= 0
= 13.9
13.8
13.7
0 2 4 6 8 10
time (min.)

Plot of TGA results for Cr(acac)3 performed
at different isothermal regions. Adapted
from B. D. Fahlman and A. R. Barron, Adv.
Mater. Optics Electron., 2000, 10, 223.

Note:Samples of iron acetylacetonate ({link]a, where M = Fe, n = 3) may
be used as a calibration standard through AH,,,, determinations before each
day of use. If the measured value of the sublimation enthalpy for Fe(acac)3
is found to differ from the literature value by more than 5%, the sample is
re-analyzed and the flow rates are optimized until an appropriate value is
obtained. Only after such a calibration is optimized should other
complexes be analyzed. It is important to note that while small amounts (<
10%) of involatile impurities will not interfere with the AH,,, analysis,
competitively volatile impurities will produce higher apparent sublimation
rates.

It is important to discuss at this point the various factors that must be
controlled in order to obtain meaningful (useful) m,,, data from TGA data.

1. The sublimation rate is independent of the amount of material used but
may exhibit some dependence on the flow rate of an inert carrier gas,
since this will affect the equilibrium concentration of the cubane in the
vapor phase. While little variation was observed we decided that for
consistency Mg, values should be derived from vacuum experiments
only.

2. The surface area of the solid in a given experiment should remain
approximately constant; otherwise the sublimation rate (i.e.,
mass/time) at different temperatures cannot be compared, since as the
relative surface area of a given crystallite decreases during the
experiment the apparent sublimation rate will also decrease. To
minimize this problem, data was taken over a small temperature ranges
(ca. 30 °C), and overall sublimation was kept low (ca. 25% mass loss
representing a surface area change of less than 15%). In experiments
where significant surface area changes occurred the values of M.yp
deviated significantly from linearity on a log(m,,,) versus 1/T plot.

3. The compound being analyzed must not decompose to any significant
degree, because the mass changes due to decomposition will cause a
reduction in the apparent m,,, value, producing erroneous results. With
a simultaneous TG/DTA system it is possible to observe exothermic
events if decomposition occurs, however the clearest indication is
shown by the mass loss versus time curves which are no longer linear
but exhibit exponential decays characteristic of first or second order
decomposition processes.

Data analysis

The basis of analyzing isothermal TGA data involves using the Clausius-
Clapeyron relation between vapor pressure (p) and temperature (T), [link],
where AH,,p is the enthalpy of sublimation and R is the gas constant (8.314
J/K.mol).

Equation:

Since msub data are obtained from TGA data, it is necessary to utilize the
Langmuir equation, [link], that relates the vapor pressure of a solid with its
sublimation rate.

Equation:

0.5
_ 2a ,
= My sub

After integrating [link] in log form, substituting in [link], and consolidating
the constants, one obtains the useful equality, [link].

Equation:
=——— 9.0522(AHsub) —
leptin el) 0.0522(AHsub ) 4 | 0.0522(AHsub 1 log 1306
T Tsub Mw

Hence, the linear slope of a log(m,,, 1/7) versus 1/T plot yields AHs,». An
example of a typical plot and the corresponding AH, value is shown in
[link]. In addition, the y intercept of such a plot provides a value for T.,p,
the calculated sublimation temperature at atmospheric pressure.

log (m.T 12)
oO

i]
=

-2
0.0022 0.0023 0.0024 0.0025

1/T (K71)

Plot of log(m,,,T 7) versus 1/T and the
determination of the AH,,, (112.6 kJ/mol)
for Fe(acac)3 (R* = 0.9989). Adapted from

B. D. Fahlman and A. R. Barron, Adv.

Mater. Optics Electron., 2000, 10, 223.

[link] lists the typical results using the TGA method for a variety of metal
B-diketonates, while [link] lists similar values obtained for gallium
chalcogenide cubane compounds.

Compound AH sub AScub ain Calculated
(kJ/mol) (J/K.mol) calc. vapor
(°C) pressure @
150 °C
(Torr)

Al(acac)3
Al(tfac)3

Al(hfac)3
Cr(acac)3
Cr(tfac)3

Cr(hfac)3
Fe(acac)3
Fe(tfac)3

Fe(hfac)s3
Co(acac)3
Co(tfac)3

Co(hfac)3

93

74

a2

91

71

46

112

96

60

138

119

73

220

192

152

216

186

134

259

243

169

311

295

200

150

111

70

148

109

69

161

121

81

170

131

90

3.261

9.715

29.120

3.328

9.910

29.511

2.781

8.340

25.021

1.059

3.319

9.132

Selected thermodynamic data for metal B-diketonate compounds
determined from thermogravimetric analysis. Data from B. D. Fahlman and

A. R. Barron, Adv. Mater. Optics Electron., 2000, 10, 223.

Compound

AHgup
(kJ/mol)

ASsub T sub
calc.
mol) (°C)

(J/K.

Calculated
vapor
pressure

@ 150°C

(Torr)
[(Me3C)GaS ]4 110 300 94 2203
[(EtMe,C)GaS ], 124 330 102 18.89
[(EtyMeC)GaS ]4 137 339 131 1.173
[(Et3C)GaS ]4 149 333 175 0.018
[(Me3C)GaSe) |, 119 305 116 3.668
[(EtMe,C)GaSe], 137 344 124 2.562
[(EtyMeC)GaSe], 147 359 136 0.815
[(EtzC)GaSe], 156 339 189 0.005

Selected thermodynamic data for gallium chalcogenide cubane compounds
determined from thermogravimetric analysis. Data from E. G. Gillan, S. G.
Bott, and A. R. Barron, Chem. Mater., 1997, 9, 3, 796.

A common method used to enhance precursor volatility and corresponding
efficacy for CVD applications is to incorporate partially ([link]b) or fully
({link]c) fluorinated ligands. As may be seen from [link] this substitution
does results in significant decrease in the AH,,, and thus increased
volatility. The observed enhancement in volatility may be rationalized either
by an increased amount of intermolecular repulsion due to the additional
lone pairs or that the reduced polarizability of fluorine (relative to
hydrogen) causes fluorinated ligands to have less intermolecular attractive
interactions.

Determination of sublimation entropy

The entropy of sublimation is readily calculated from the AH,,,, and the
calculated T.,, data, [link].

Equation:

ASwup 7 AH up
T

sub

[link] and [link] show typical values for metal B-diketonate compounds and
gallium chalcogenide cubane compounds, respectively. The range observed
for gallium chalcogenide cubane compounds (AS, = 330 +20 J/K.mol) is
slightly larger than values reported for the metal $-diketonates compounds
(ASgyp = 130 - 330 J/K.mol) and organic compounds (100 - 200 J/K.mol),
as would be expected for a transformation giving translational and internal
degrees of freedom. For any particular chalcogenide, i.e., [((R)GaS],, the
lowest AS,,,, are observed for the Me3C derivatives, and the largest AS,
for the EtpMeC derivatives, see [link]. This is in line with the relative
increase in the modes of freedom for the alkyl groups in the absence of
crystal packing forces.

Determination of vapor pressure

While the sublimation temperature is an important parameter to determine
the suitability of a potential precursor compounds for CVD, it is often
preferable to express a compound's volatility in terms of its vapor pressure.
However, while it is relatively straightforward to determine the vapor
pressure of a liquid or gas, measurements of solids are difficult (e.g., use of
the isoteniscopic method) and few laboratories are equipped to perform
such experiments. Given that TGA apparatus are increasingly accessible, it
would therefore be desirable to have a simple method for vapor pressure
determination that can be accomplished on a TGA.

Substitution of [link] into [link] allows for the calculation of the vapor
pressure (p) as a function of temperature (T). For example, [link] shows the
calculated temperature dependence of the vapor pressure for [((Me3C)GaS]j.
The calculated vapor pressures at 150 °C for metal 6-diketonates
compounds and gallium chalcogenide cubane compounds are given in [link]
and [link].

300

N N
(o) oO
oO oO

Vapor Pressure (Tom
a
(ao)

360 380 400 420 440 460 £480
Temperature (K)

A plot of calculated vapor pressure (Torr)
against temperature (K) for [((Me3C)GaS],.
Adapted from E. G. Gillan, S. G. Bott, and
A. R. Barron, Chem. Mater., 1997, 9, 3, 796.

The TGA approach to show reasonable agreement with previous
measurements. For example, while the value calculated for Fe(acac)3 (2.78
Torr @ 113 °C) is slightly higher than that measured directly by the
isoteniscopic method (0.53 Torr @ 113 °C); however, it should be noted
that measurements using the sublimation bulb method obtained values
much lower (8 x 10° Torr @ 113 °C). The TGA method offers a suitable
alternative to conventional (direct) measurements of vapor pressure.

Bibliography

e P. W. Atkins, Physical Chemistry, 5th ed., W. H. Freeman, New York
(1994).
e G. Beech and R. M. Lintonbon, Thermochim. Acta, 1971, 3, 97.

B. D. Fahlman and A. R. Barron, Adv. Mater. Optics Electron., 2000,
10, 223.

E. G. Gillan, S. G. Bott, and A. R. Barron, Chem. Mater., 1997, 9, 3,
796.

J. O. Hill and J. P. Murray, Rev. Inorg. Chem., 1993, 13, 125.

J.P. Murray, K. J. Cavell and J. O. Hill, Thermochim. Acta, 1980, 36,
97.

M. A. V. Ribeiro da Silva and M. L. C. C. H. Ferrao, J. Chem.
Thermodyn., 1994, 26, 315.

R. Sabbah, D. Tabet, S. Belaadi, Thermochim. Acta, 1994, 247, 193.
L. A. Torres-Gomez, G. Barreiro-Rodriquez, and A. Galarza-
Mondragon, Thermochim. Acta, 1988, 124, 229.

Phosphine and Arsine
Because of their use in metal organic chemical vapor deposition (MOCVD)

of 13-15 (III-V) semiconductor compounds phosphine (PH3) and arsine
(AsHs3) are prepared on an industrial scale.

Synthesis

Phosphine (PH3) is prepared by the reaction of elemental phosphorus (P,)
with water, [link]. Ultra pure phosphine that is used by the electronics
industry is prepared by the thermal disproportionation of phosphorous acid,
[link].

Equation:

2P,+ 12H,O > 5PH; + 3H,PO,
Equation:
4H;PO; > PH; + 3H,PO,
Arsine can be prepared by the reduction of the chloride, [link] or [link]. The

corresponding syntheses can also be used for stibine and bismuthine.
Equation:

4 AsCl, + 3LiAIH, > 4 AsH, + 3 LiAICI,
Equation:

4 AsCl,+ 3NaBH, > 4 AsH,; + 3 NaCl + 3 BCI

The hydrolysis of calcium phosphide or arsenide can also generate the
trihydrides.

Structure

The phosphorus in phosphine adopts sp? hybridization, and thus phosphine
has an umbrella structure ({link]a) due to the stereochemically active lone
pair. The barrier to inversion of the umbrella (E, = 155 kJ/mol) is much
higher than that in ammonia (E, = 24 kJ/mol). Putting this difference in
context, ammonia’s inversion rate is 10!! while that of phosphine is 10°. As
a consequence it is possible to isolate chiral organophosphines (PRR'R").
Arsine adopts the analogous structure ([link]b).

D 1.42 A A. 1.519A
wep H ALY
H 7H

H 7H
93.5° 91.8°
(a) (b)

The structures of (a) phosphine
and (b) arsine.

Reactions

Phosphine is only slightly soluble in water (31.2 mg/100 mL) but it is
readily soluble in non-polar solvents. Phosphine acts as neither an acid nor
a base in water; however, proton exchange proceeds via the phosphonium
ion (PH,°) in acidic solutions and via PH>’ at high pH, with equilibrium
constants K, = 4 x 10°°8 and K, = 41.6 x 10°2°, respectively.

Arsine has similar solubility in water to that of phosphine (i.e., 70 mg/100
mL), and AsH3 is generally considered non-basic, but it can be protonated
by superacids to give isolable salts of ASH”. Arsine is readily oxidized in
air, [link].
Equation:

2 AsH,; + 30, > As,O, + 3H,O

Arsine will react violently in presence of strong oxidizing agents, such as
potassium permanganate, sodium hypochlorite or nitric acid. Arsine
decomposes to its constituent elements upon heating to 250 - 300 °C.

Gutzeit test

The Gutzeit test is the characteristic test for arsenic and involves the
reaction of arsine with Ag*. Arsine is generated by reduction of aqueous
arsenic compounds, typically arsenites, with Zn in the presence of H»SOy,.
The evolved gaseous AsH3 is then exposed to silver nitrate either as powder
or as a Solution. With solid AgNO3, AsHs3 reacts to produce yellow
Ag,AsNOs3, while with a solution of AgNO3 black Ag3As is formed.

Hazards

Pure phosphine is odorless, but technical grade phosphine has a highly
unpleasant odor like garlic or rotting fish, due to the presence of substituted
phosphine and diphosphine (P)H,). The presence of PH, also causes
spontaneous combustion in air. Phosphine is highly toxic; symptoms
include pain in the chest, a sensation of coldness, vertigo, shortness of
breath, and at higher concentrations lung damage, convulsions and death.
The recommended limit (RL) is 0.3 ppm.

Arsine is a colorless odorless gas that is highly toxic by inhalation. Owing
to oxidation by air it is possible to smell a slight, garlic-like scent when
arsine is present at about 0.5 ppm. Arsine attacks hemoglobin in the red
blood cells, causing them to be destroyed by the body. Further damage is
caused to the kidney and liver. Exposure to arsine concentrations of 250
ppm is rapidly fatal: concentrations of 25 — 30 ppm are fatal for 30 min
exposure, and concentrations of 10 ppm can be fatal at longer exposure
times. Symptoms of poisoning appear after exposure to concentrations of
0.5 ppm and the recommended limit (RL) is as low as 0.05 ppm.

Bibliography

e R. Minkwitz, A. Kornath, W. Sawodny, and H. Hartner, Z. Anorg. Allg.
Chem., 1994, 620, 753.

Mechanism of the Metal Organic Chemical Vapor Deposition of Gallium
Arsenide

Introduction

Preparation of epitaxial thin films of III-V (13-15) compound
semiconductors (notably GaAs) for applications in advanced electronic
devices became a realistic technology through the development of metal
organic chemical vapor deposition (MOCVD) processes and techniques.
The processes mainly involves the thermal decomposition of metal alkyls
and/or metal hydrides.

In 1968 Manasevit at the Rockwell Corporation was the first to publish on
MOCVD for the epitaxial growth of GaAs. This followed his pioneering
work in 1963 with the epitaxial growth of silicon on sapphire. The first
publication used triethylgallium [Ga(CH»CH3)3] and arsine (AsH3) in an
open tube with hydrogen as the carrier gas. Manasevit actually coined the
phrase MOCVD and since this seminal work there have been numerous
attempts to improve and expand MOCVD for the fabrication of GaAs.

Several processes, partly in series, partly in parallel take place during the
growth by CVD. They are presented schematically in [link]. The relative
importance of each of them depends on the chemical nature of the species
involved and the design of the reactor used. The actual growth rate is
determined by the slowest process in the series of events needed to come to
deposition.

[missing_resource: GaAs Fig 1.jpg]

Schematic representation of
the fundamental transport and
reaction steps underlying
MOCVD. Adapted from K. F.
Jensen and W. Kern, in Thin
Film Processes II, Eds. J. L.
Vossen and W. Kern,

Academic Press, New York
(1991).

Conventionally, the metal organic chemical vapor deposition (MOCVD)
growth of GaAs involves the pyrolysis of a vapor phase mixture of arsine
and, most commonly, trimethylgallium [Ga(CH3)3, TMG] and
triethylgallium [Ga(CH»CH3)3, TEG]. Traditionally, growth is carried out in
a cold-wall quartz reactor in flowing H> at atmospheric or low pressure.
The substrate is heated to temperatures 400 - 800 °C, typically by RF
heating of a graphite susceptor. Transport of the metal-organics to the
growth zone is achieved by bubbling a carrier gas (e.g., H>) through the
liquid sources that are in held temperature-controlled bubblers.

Reaction mechanism

While the overall reaction (where R = CH3 or CH»CH3) can be described
by [link].
Equation:

R,Ga + AsH,; — GaAs + 3 RH

The nature of the reaction is much more complex. From early studies it was
thought that free Ga atoms are formed by pyrolysis of TMG and As,
molecules are formed by pyrolysis of AsH3 and these species recombine on
the substrate surface in an irreversible reaction to form GaAs.

Although a Lewis acid-base complex formed between TMG and AsHs3 is
possible, it is now known that if there is any intermediate reaction between
the TMG and AsHs, the product is unstable. However, early work indicated
that free GaAs molecules resulted from the decomposition of a TMG.AsH3
intermediate and that the heated surface contributed to the reaction. It was
subsequently suggested that the reaction occurs by separate pyrolysis of the
reactants and a combination of individual Ga and As atoms at the surface or
just above it. Finally, evidence has also been found for TMG pyrolysis

followed by diffusion through a boundary layer and for AsH3 pyrolysis
catalyzed by the GaAs surface.

There are several different kinds of potential reactions occuring in the CVD
reaction chamber, namely, ligand dissociation, ligand association, ,
reductive elimination, oxidative addition, B-hydride elimination, etc. Some
of them are listed in the following equations:

Equation:

Ga(CH;); —> Ga(CH;), + CH, (ligand dissociation)

Equation:

Ga(CH;), — Ga + CH,-CH, (reductive elimination)

Equation:

Ga(CH;), —> Ga(CH;) + CH,-CH, (reductive elimination)

Equation:

CH,+H — CH, (radical recombination)

Equation:

Ga-CH,-CH, —> Ga-H + H,C=CH, (6-hydride elimination)

Equation:

Ga(CH;), + AsR,; — (CH;),Ga-AsR, (ligand association)

Using ALE studies as insight for MOCVD

Given the stepwise and presumably simplified mechanism for atomic layer
epitaxy (ALE) growth of GaAs, a number of mechanistic studies have been
undertaken of ALE using TMG and AsH;3 to provide insight into the
comparable MOCVD reactions. Nishizawa and Kurabayashi proposed that
a CH3-terminated GaAs surface inhibits further heterogeneous
decomposition of TMG and self-limits the growth rate to one
monolayer/cycle. While, X-ray photoelectron spectroscopy (XPS) studies
showed that no carbon was observed on a GaAs surface reacted with TMG.
Furthermore, the same self-limiting growth was seen in in ALE using a
metalorganic molecular beam epitaxy (MOMBE) with TMG and AsHsz. It
was reported that a transient surface reconstruction is observable by
reflection high-energy electron diffraction (RHEED) during the ALE of
GaAs in MOMBE. It was suggested that this structure is caused by CH3-
termination and the self-limitation of the growth rate is attributed to this
structure. However, measurement of the desorption of CH3 by means of a
combination of pulsed molecular beams and time-resolved mass
spectrometry, indicates that CH3 desorption is too fast to attribute the self-
limitation to the CH3-terminated surface. Subsequently, investigations of
the pyrolysis of TMG on a (100)GaAs surface by the surface photo-
absorption method (SPA) allowed for the direct observation of CH3
desorption from a GaAs surface reacted with TMG. From the measured
CH3 desorption kinetics, it was shown that the CH3-terminated surfaces
causes the self-limitation of the growth rate in ALE because the excess
TMG cannot adsorb.

All this research helped people to visualize the real reaction mechanism in
the formation of GaAs by MOCVD methods, in which the decomposition,
diffusion and surface reaction interact with each other and result in a much
more complicated reaction mechanism.

Gas phase reaction: pyrolysis of TMG and AsH3

In the TMG/Hp system, there is almost no reactions at a temperature below
450 °C, whereas the reaction of TMG with H> almost completely changed
into CH, and Ga at a temperature above 600 °C, [link].

Equation:

Ga(CH;); + 7/3H, > Ga + 3CH,

As for the AsH3 decomposition, without any deposition of Ga or GaAs in
the reactor, the pyrolysis of AsH3 proceeded barely at a temperature below
600 °C, however, it proceeded nearly completely at a temperature above
750 °C. In the AsH3/H> system with the TMG introduced previously, the
decomposition of AsH3 was largely enhanced even at a temperature below
600 °C. The decomposition of AsH3 seems to be affected sensitively by the
deposited GaAs or Ga. This phenomenon may be concluded to be caused by
the catalytic action by GaAs or Ga. The reaction at a temperature below 600
°C can be described as shown in [link], but at a temperature above 600 °C,
pyrolysis of AsH3 can occur even without GaAs or Ga, [link].

Equation:

GaAs
ASH3 (ag) = AS (ag) + eae

Equation:

AsH, — As + 37/,H,

Adsorption and surface reactions

From the temperature dependent measurements of the desorption spectrum
from a surface on which TMG was supplied, it was estimated that the
surface-adsorbed species was Ga at the high temperature region of T,,, >
500 °C, GaCHs3 at the range of 350 °C < T,,, < 500 °C, and Ga(CH3), and
Ga(CH)3)3 at the range of T.,, < 350 °C. The reactions, where (ad) means
the adsorbed state of the molecules, are:

Equation:

< 350 °C)

sub

Equation:

Ga(CHs)3(.) > ~Ga(CHg)5'(aa) (Ty, < 350 °C)
Equation:

Ga(CHs)3(.) > GaCHy,,3, + 2 CH, (350°C <T,,, < 500 °C)
Equation:

Ga(CH)3(.) > Gaia + 3 CH, (Tu, > 500 °C)

When AsHs3 is supplied, the reactions with these adsorbates are:
Equation:

Ga(CH;), + AsH, — no reaction (Tsu, < 350 °C)
Equation:
Ga(CH3)3 (a4) + ASH; —> no reaction (Tun < 350 °C )

Equation:

GaCHy (aq) + ASH; -> GaAs + CH, + H, (350<T,,, < 500°C)
Equation:

Gag) + ASH; — GaAs + 7/,H, (T,,,, > 500 °C)

It was observed that there is no growth in the range of T,,,, < 350 °C, i.e.,
Ga(CH3)9‘(aq) and Ga(CH3)3/aq) do not react with AsH3 in the TMG-AsH3
system. Monomolecular layer growth is limited by the formation of GaCH3
and its reaction with AsH3.

Overall reaction pathway

At lower temperature (350 - 500 °C), equivalently low energy, TMG
decompose in the gas phase to Ga(CH3)5 and methyl radical, [Link].
Equation:

Ga(CH;), —> Ga(CH;), + CH, (low energy, gas phase)

After the first ligand dissociation, there are two different pathways, in the
first, the Ga(CH3) keeps decomposing into GaCH3 and another methyl
group when it is at the gas-substrate interface, [link], and then further
decomposes into free gallium atoms on the substrate surface, [link]. In the
second reaction, the Ga(CH3). decomposes directly into Ga and CH3-CH3
by reductive elimination, [link].

Equation:

Ga(CH;), —> GaCH, + 2 CH, (gas/surface)
Equation:

GaCH, — Ga + CH, (surface)
Equation:

Ga(CH;), — Ga + H,C-CH, (reductive elimination)

At high temperature (> 500 °C), the TMG decomposes into Ga(CH3) and
two methyl groups instead of the step-wise decomposition at lower
temperature, [link], and the Ga(CH3) further decomposes into free Ga atoms
at the substrate surface, [link].

Equation:

Ga(CH;), — GaCH, + 2 CH, (high temperature, gas phase)

Equation:

GaCH, — Ga + CH,

The decomposition of AsH3 forms an “arsenic cloud” in the reaction
chamber. The decomposition is also step-wise:
Equation:

AsH,; — AsH,

Equation:

AsH, — HAs + H (surface)

Equation:

HAs — As + H (surface)

The methyl groups in the surface Ga(CH3) molecules are removed by the
formation of methane with atomic hydrogen from the decomposition of
AsHs3, [Link].

Equation:

H + CH, — CH, (surface)

Kinetics for other systems

Investigations have been reported for the mechanism of the growth of GaAs
using triethylgallium [Ga(CH»CH3)3, TEG] and TMG with trimethylarsene
[As(CH3)3, TMA], triethylarsene [As(CH»CH3)3, TEA], tert-butylarsine

{[(CH3)3C]AsH», TBA}, and phenylarsine [(CgH,)AsH>]. The experiments

were conducted in a MOCVD reactor equipped with a recording
microbalance for in-situ growth rate measurements. For example, the
kinetics of the growth of GaAs were investigated by measuring growth rate
as a function of temperature using the microbalance reactor while holding
the partial pressure of gallium precursor (e.g., TMG) and arsenic precursor
[e.g., As(CH3)3] constant at 0.01 and 0.05 Torr, respectively. Three different
flow rates were used to determine the influence of the gas residence time.

The growth rate of GaAs with TMG and As(CH»CH3)3 is higher as
compared with the growth from TMG and As(CHs3)3 because of the lower
thermal stability of As(CH»CH3)3 than As(CH3)3. Both of the two growth
rates showed a strong dependence on the residence time.

Similarly, the kinetic behaviors of the TMG/TBA and TEG/TBA system
were investigated under the same conditions as the TMA and TEA studies.
There are two distinct regions of growth. For TMG/TBA, the deposition
rate is independent at low temperature and in the intermediate temperature
(around 600 °C) the dependence of the growth rate on the total flow rate is
significant. This means that the growth at the lower temperature is
controlled by surface reactions. The TEG/TBA system showed a similar
behavior except that the maximum growth rate occurs around 450 °C while
it is around 750 °C for TMG/TBA system. Also, the growth of
TMG/(CgH;)AsH> was studied on the same conditions as for the
Me3Ga/‘BuAsH, system. It was reported that the difference in the growth
rate at various flow rates was related to a combination of parasitic reactions
and depletion effects from deposition. From the comparison of the data, it is
deduced that the effect of parasitic reactions is slightly smaller for
(CgHs)AsH> than for TBA.

Two possible mechanisms for the dependence of growth rate on flow rate
were proposed. The first, mass-transfer limitation was thought to be
unlikely because of the high diffusivity of the gallium precursors at 1 Torr
(ca. 350 cm?/s). The second, also the more likely explanation for the
observed growth-rate dependence on flow rates is gas-phase depletion cause
by the parasitic reactions. Since the growth efficiency is high (41% at 700
°C), the loss of precursor from the gas phase will directly affect the growth
rate. It was evidenced by the differences in the growth rates between split

and combined feed streams. The growth rate is lower when the reagents are
combined upstream of the reactor than when they are combined inside the
reactor (split stream). It is suggested that the experimental observations can
be explained by a model based on the reversible formation of an adduct and
the decomposition of this adduct to useless polymeric material competing
with the growth of GaAs. It can be written in the form shown in [link]
where ky and k, are the forward and reverse rate constants for adduct
formation, respectively, kg is the rate constant for the irreversible
decomposition of the adduct to polymer, and k, is the surface reaction rate
constant for the growth of GaAs. It is obvious that each step involves
several elementary reactions, but there were insufficient data to provide any
more detail.

Equation:

ky d
GaAs <— organometallic precursors === adduct — polymeric deposits

Bibliography

e T.H. Chiu, Appl. Phys. Lett., 1989, 55, 1244.

e H. Ishii, H. Ohno, K. Matsuzaki and H. Hasegawa, J. Crys. Growth,
1989, 95, 132.

e K. F. Jensen and W. Kern, in Thin Film Processes II, Eds. J. L. Vossen
and W. Kern, Academic Press, New York (1991).

e N. Kobayashi, Y. Yamauchi, and Y. Horikoshi, J. Crys. Growth, 1991,
115, 353.

e K. Kodama, M. Ozeki, K. Mochizuki, and N. Ohtsuka, Appl. Phys.
Lett., 1989, 54, 656.

e M.R. Leys and H. Veenvliet, J. Crys. Growth, 1981, 55, 145.

e U. Memmert and M. L. Yu, Appl. Phys. Lett., 1990, 56,1883.

e J. Nishizawa and T. Kurabayashi, J. Crys. Growth. 1988, 93, 132.

e T. R. Omstead and K. F. Jensen, Chem. Mater., 1990, 2, 39.

e D.J. Schyer and M. A. Ring, J. Electrochem. Soc., 1977, 124, 569.

e Watanabe, T. Isu, M. Hata, T. Kamijoh, and Y. Katayama, Japan. J.
Appl. Phys., 1989, 28, L1080.

e Y. Zhang, Th. Beuermann, and M. Stuke, Appl. Phys. B, 1989, 48, 97.

e Y. Zhang, W. M. Cleaver, M. Stuke, and A. R. Barron, Appl. Phys. A,
1992, 55, 261.

Chemical Vapor Deposition of Silica Thin Films

General considerations

Before describing individual chemical vapor deposition (CVD) systems for the deposition of
silica thin films, it is worth outlining general considerations to be taken into account with regard
to the growth by CVD of any insulating film: the type of CVD method, deposition variables,
and limitations of the precursor.

Deposition methods

In regard to the CVD of insulating films in general, and silica films in particular, three general
reactors are presently used: atmospheric pressure CVD (APCVD), low and medium temperature
low pressure CVD (LPCVD), and plasma-enhanced CVD (PECVD). LPCVD is often further
divided into low and high temperatures.

APCVD systems allow for high throughput and even continuous operation, while LPCVD
provides for superior conformal step coverage and better film homogeneity. PECVD has been
traditionally used where low temperatures are required, however, film quality is often poor. As
compared to PECVD, photo-assisted CVD has the additional advantage of highly selective
deposition, although it has been little used in commercial systems. [link] summarizes the
advantages and disadvantages of each type of CVD system commercially used for SiO, films.

Atmospheric Low Medium Plasma
pressure temperature temperature enhanced
CVD LPCVD LPCVD CVD
a 300 - 500 300 - 500 500 - 900 100 - 350
Throughput high high high low
aap poor poor conformal poor
coverage
igs : good good excellent poor
properties
iisés passivation, passivation, ‘cenlaiian passivation,

insulation insulation insulation

Comparison of different deposition methods for SiO» thin films.

Deposition variables

The requirements of CVD films for electronic device applications have become increasingly
more stringent as device sizes are continually reduced. Film thickness must be uniform across
an entire wafer, i.e., better than +1%. The structure of the film and its composition must be
controlled and reproducible, both on a single wafer, as well as between wafer samples. It is also
desirable that the process is safe, inexpensive, and easily automated.

A number of variables determine the quality and rate of film growth for any material. In general,
the deposition rate increases with increased temperature and follows the Arrhenius equation,
[link], where R is the deposition rate, E, is the activation energy, T is the temperature (K), A is
the frequency factor, and k is Boltzmann's constant (1.381 x 10°23 J/K).

Equation:

R = A exp(-E,/kT)

At the high temperatures the rate of deposition becomes mass transport limited. Meaning, the
rate of surface reaction is faster than the rate at which precursors are transported to the surface.
In multiple source systems, the film growth rate is dependent on the vapor phase concentration
(or partial pressure) of each of the reactants, but in certain cases the ratio of reactants is also
important, e.g., the SiH,/O» growth of SiO». Surface catalyzed reactions can also alter the
deposition rate. Such as the non-linear dependence of the deposition rate of SiO on the partial
pressure of Si(OEt),4. Gas depletion may also be significant requiring either a thermal ramp in
the chamber and/or special reactor designs. The necessary incorporation of dopants usually
lowers deposition rates, due to competitive surface binding.

For the applications of insulating materials as isolation layers, an important consideration is step
coverage: whether a coating is uniform with respect to the surface. [link]a shows a schematic of
a completely uniform or conformal step coverage of a trench (such as occurs between isolated
devices) where the film thickness along the walls is the same as the film thickness at the bottom
of the step. Uniform step coverage results when reactants or reactive intermediates are able to
migrate rapidly along the surface before reacting. When the reactants adsorb and react without
significant surface migration, deposition is dependent on the mean free path of the gas. [link]b
shows an example of minimal surface migration and a short mean free path. For SiO» film
growth LPCVD has highly uniform coverage ([link]a) and PECVD poor step coverage ([link]b).

film ———~>

substrate ——>

(a) (b)

Step coverage of deposited films with (a) uniform
coverage resulting from rapid surface migration and (b)
nonconformal step coverage due to no surface migration.

Precursor considerations

The general requirements for any CVD precursor have been adequately reviewed elsewhere, and
will not be covered here. However, many of the gases and organometallics used to deposit
dielectric films are hazardous. The safety problems are more severe for LPCVD because the
process often uses no diluent gas such as argon or nitrogen. [link] lists the boiling point and
hazards of common inorganic and organometallic precursor sources for CVD of SiO» and doped
silica. Many of the precursors react with air to form solid products, thus leaks can cause
particles to form in the chamber and gas lines.

Gas Formula Bpt (°C) Hazard
ammonia NH3 -33.35 toxic, corrosive
argon Ar -185.7 inert

arsine AsH3 -55 toxic

diborane BoHg -92.5 toxic, flammable
dichlorosilane SiCl)H> 8.3 toxic, flammable
hydrogen Hy -252.8 flammable

nitrogen N> -209.86 inert

nitrous oxide N»O -88.5 oxidizer

oxygen Oz -182.962 oxidizer

phosphine PH3 -87.7 toxic, P2H, impurities, flammable
silane SiH, -111.8 flammable, toxic

Physical and hazard properties of common gaseous sources for CVD of dielectric materials.

In principle, the deposition of a SiO», or silica, thin film by CVD requires two chemical sources:
the element (or elements) in question, and an oxygen source. While dioxygen (O3) is suitable for
many applications, its reactions may be too fast or too slow for optimum film growth, requiring
that alternative oxygen sources be used, e.g., nitrous oxide (N»O) and ozone (O3). A common
non-oxidizing oxygen source is water. A more advantageous approach is to incorporate oxygen
into the ligand environment of the precursor, and endeavor to preserve such an interaction intact
from the source molecule into the ultimate film; such a source is often termed a "single-source"
precursor.

CVD silica (SiO,)

The processing sequence for silicon dioxide (SiO») used depends on its specific use. CVD
processes for SiO» films can be characterized by either the chemical reaction type, the growth
pressure, or the deposition temperature. The choice of route is often dictated by requirements of
the thermal stability of the substrate or the conformality. [link] summarizes selected properties
of SiO» grown by various CVD methods, in comparison to that of thermally grown silica. In
general, silica grown at high temperatures resemble thermally grown “native” SiO». However,
the use of aluminum metallization requires low temperature deposition of silica.

SiCI,H>

Deposition Plasma O, Si(OEt), +N,0 Thermal
ae 200 450 700 900 1000
Composition SiO, 9(H) SiO»(H) SiO» SiO2(Cl) SiO»

= ee es ae conformal conformal conformal
Thermal loses H densifies stable loses Cl stable

stability

Refractive

1.47 1.44 1.46 1.46 1.46
idex

Dielectric 49 43 4.0 4.0 Be,
constant

Comparison of physical properties of SiO) grown by commercial CVD methods.

CVD from hydrides

The most widely used method for SiO, thin film CVD is the oxidation of silane (SiH), first
developed in 1967 for APCVD. Nonetheless, LPCVD systems have since become increasingly
employed, and exceptionally high growth rates (30,000 A/min) have been obtained by the use of
rapid thermal CVD.

The chemical reaction for SiO» deposition from SiH, is:
Equation:

SiH, + O, > SiO, + 2H,

At high oxygen partial pressures an alternative reaction occurs, resulting in the formation of
water.
Equation:

SiH, + 20, > SiO, + 2H,O

While these reactions appears simple, the detailed mechanism involves a complex branching-
chain sequence of reactions. The apparent activation energy is low (< 41 kJ/mol) as a
consequence of its heterogeneous nature, and involves both surface adsorption and surface
catalysis.

Nitrous oxide (N2O) can be used as an alternative oxygen source to O2, according to the overall
reaction, [link].
Equation:

SiH, + 2N,O > SiO, + 2H, + 2N,

A simple kinetic scheme has been developed to explain many of the observed aspects of SiHy-
NO growth. It was suggested that the reaction is initiated by decomposition of NO, [link],
generating an oxygen radical which can abstract hydrogen from silane forming a hydroxyl
radical, [link], that can react further with silane, [link].

Equation:

N,0 > N, +0

Equation:

SiH, + O > SiH, + OH

Equation:

SiH, + OH > SiH, + H,O

Evidence for the reaction of the OH radical to form water is the formation of a small quantity of
water observed during the oxidation of SiH4. Silyl radicals are oxidized by N2O to form siloxy
radicals, [link], which provide a suitable propagation step, [link].

Equation:

SiH, + N,O > SiH,O +N,

Equation:

SiH,O + SiH, > SiH,OH + SiH,

It has been proposed that the silanol (SiH3OH) is the penultimate film precursor.

The SiHy-O> and SiH,-N>O routes to SiO, thin films are perhaps the most widely studied
photochemical CVD system of all dielectrics. Photo-C VD of SiO, provides a suitable route to
deposition at low substrate temperatures, thereby avoiding potential thermal effects of wafer
warpage and deleterious dopant redistribution. In addition, unlike other low temperature
methods such as APCVD and PECVD, photo-CVD often provides good purity of films.

A summary of common silane CVD systems is given in [link].

Oxygen Carrier gas CVD Deposition Growth rate
source (diluent) method temp. (°C) (A/min)
O> N> APCVD 350 - 475 100 - 14,000

Oz Ar LPCVD 100 - 550 100 - 30,000

O» At/N> LPCVD 25 - 500 10 - 450

O, Ar PECVD 25 - 200 200 - 900
N,O N> APCVD 490 - 690 200 - 1,200
N,O N> LPCVD 700 - 860 ca. 50

N,O N> LPCVD 25 - 350 7 - 180
N,O Ar PECVD 100 - 200 80 - 800

Precursors and deposition conditions for SiO, CVD using silane (SiH).

CVD from halides

The most widely used process of the high temperature growth of SiO) by LPCVD involves the
N>O oxidation of dichlorosilane, SiC] )Hb, [link].
Equation:

SiCL,H, + 2N,O > SiO, + 2HCI + 2N,

Deposition at 900 - 915 °C allows for growth of SiO, films at ca. 120 A/min; however, these
films are contaminated with Cl. Addition of small amounts of O» is necessary to remove the
chlorine.

While PECVD has been employed utility halide precursors, the ability of small quantities of
fluorine to improve the electrical properties of SiO) has prompted investigation of the use of
SiF,4 as a suitable source.

CVD from tetraethoxylsilane (TEOS)

The first CVD process to be introduced into semiconductor technology in 1961 was that
involving the pyrolysis of tetraethoxysilane, Si(OEt)4 (commonly called TEOS from
tetraethylorthosilicate). Deposition occurs at an optimum temperature around 750 °C. However,
under LPCVD conditions, the growth temperature can be significantly lowered (> 600 °C). The
high temperature growth of SiO» from TEOS involves no external oxygen source. Dissociative
adsorption studies indicate that decomposition of the TEOS-derived surface bound di- and tri-
ethoxysiloxanes is the direct source of the ethylene.

PECVD significantly lowers deposition temperatures using TEOS, but requires the addition of
O, to remove carbon contamination, via the formation of gaseous CO and CO», which are

subsequently not incorporated within the film. Although deposition as low as 100 °C may be
obtained, the film resistivity increases by three orders of magnitude by depositing at 200 °C;
being 10!° Q.cm, with a breakdown strength of 7 x 10° V/cm.

Addition of O) for APCVD growth does not decrease the deposition temperature, however, if
ozone (O3) is used as the oxidation source, deposition temperatures as low as 300 °C may be
obtained for uniform crack-free films. It has been postulated that the ozone traps the TEOS
molecule on the surface as it reacts with the ethoxy substituent, providing a lower energy
pathway (TEOS-O3 @ 55 kJ/mol versus TEOS-O2 @ 230 kJ/mol and TEOS only @ 190
kJ/mol).

There are significant advantages of the TEOS/O3 system, for example the superior step
coverage it provides. Furthermore, films have low stress and low particle contamination. On this
basis the TEOS/O3 system has become widely used for silica, as well as silicate glasses.

CVD from other organosilicon precursors

A wide range of alternative silicon sources has been investigated, especially with regard to
either lower temperature deposition and/or precursors with greater ambient stability.

Diethylsilane (Et)SiH>), 1,4-dislabutane (DBS, H3SiCH CH >SiHs3), 2,4,6,8-
tetramethylcyclotetrasiloxane (TMCTS, [link]a, where R = CH3), and 2,4,6,8-
tetraethylcyclotetrasiloxane (TECTS, [link]a, where R = CjHs), have been used in conjunction
with O, over deposition temperatures of 100 - 600 °C, depending on the precursor. Diacetoxydi-
tert-butyl silane (DADBS, [link]b) has been used without additional oxidation sources. High
quality silicon oxide has been grown at 300 °C by APCVD using the amido precursor,
Si(NMe»)4 ([link]c).

H
R, |
“Si H i
a = ZN
(0) SimR HC Oe (CH3).N
| : gy:z OC(CHs)s \ gyn! N(CH3)2
Re Si fe) r "* OC(CH3)3 7) N(CH)
rr ia ne H3C. Pasi (CH3)2N
% II
R O
(a) (b) (c)
H H
‘si—_ o— Si
HL SK ! , |
sors |
Si-|-O— Si
? St i oS ik
ly 7
Si——0 “Si
a” \
H

Alternative organometallic silicon sources that have been
investigated for the growth of silica thin films.

An interesting concept has been to preform the -Si-O-Si- framework in the precursor. In this
regard, the novel precursor Tg-hydridospherosiloxane (HgSigO1p, [link]d) gives smooth
amorphous stoichiometric SiO» at 450 - 525 °C by LPCVD. The decomposition mechanism in
the presence of added oxygen involves the loss of water, [link]. IR studies indicate that the Si-O-
Si bonds are preserved during deposition. While films are of high quality, the present synthesis
of HgSigO4> is of low yield (ca. 21%), making it currently impractical for large scale processing.
Equation:

H,Si0,. + 40, > 8SiO, + 4H,O

CVD silicate glasses

Borosilicate glasses (BSG), phosphosilicate glasses (PSG) and borophosphosilicate glasses
(BPSG) are frequently used as insulating layers separating conducting layers. These glasses
have lower intrinsic stress, lower melting temperatures and better dielectric properties than SiOz
itself. PSG and BPSG have the added property of gettering and immobilizing dopants.
Particularly important is the gettering of sodium ions, which are a source of interface traps. The
low temperature molten properties of BSG, PSG, and BPSG glasses allow for the smoothing of
the device topography by viscous thermal fusion to convert abrupt steps to more gradually
tapered steps ([link]a) as well as planarization of complex topologies ([link]b), enabling
deposition of continuous metal layers. This process is commonly called P-glass flow. The boron

and phosphorous contents of the silicate glasses vary, depending on the application, typically
being from 2 to 8 weight per cent.

BPSG

metallization

Si-substrate

(a) (b)

Schematic cross section of BPSG as deposited (a) and after
annealing (b), showing the flow causing a decrease in the angle
of the BPSG going over the step.

The advantage of BPSG over PSG is that flow occurs over the temperature range of 750 - 950
°C, depending on the relative P and B content (as opposed to 950 - 1110 °C for PSG). Lowering
of the flow temperature is required to minimize dopant migration in VLSI devices. Conversely,
the disadvantages of BPSG versus PSG include the formation of bubbles of volatile
phosphorous oxides and crystallites of boron-rich phases. If, however, the dopant concentration
is controlled, these effects can be minimized.

Arsenosilicates (AsSG) were employed originally in silicon device technology as an arsenic
dopant source for planar substrates prior to the advent of large scale ion implantation which has
largely removed the need for AsSG in doping applications. However, with ULSI silicon circuit
fabrication, the requirement for doping of deep trenches (inaccessible to ion implantation) has
witnessed the re-emergence of interest in AsSG films.

The CVD growth of silicate glasses follows that of SiO, with SiH, and TEOS being the most

commonly employed silicon precursors. A summary of common CVD precursor systems for
silicate glasses is given in [link].

CVD Deposition temp.

Precursors “aethod (°C) Applications
SiH,/B>Hg APCVD 300 - 450 good step
coverage

SiH,/B>H, LPCVD 350 - 400 :

SiH,/PH; APCVD 300 - 450 :

SiH,/PH3 LPCVD 350 - 400 flow glass
SiH4/B>H¢/PH3 APCVD 300 - 450 -
SiH4/B>H¢/PH3 LPCVD 350 - 400 -
SiH4/AsH3 APCVD 500 - 700 -
TEOS/B(OMe)3 APCVD 650 - 730 ee
source
TEOS/B(OMe)3 LPCVD 500 - 750 trench filling
TEOS/B(OEt)3 APCVD 475 - 800 oon
source
TEOS/B(OEt)3 LPCVD 500 - 750 eon
source
TEOS/PH3 LPCVD 650 flow glass
TEOS/O=P(OMe)3 APCVD 300 - 800 flow glass
diffusion
TEOS/P(OMe)3 LPCVD 500 - 750
source
TEOS/O=P(OMe)3 LPCVD 500 - 800 flow glass
TEOS/B(OMe):3/PH3 LPCVD 620 - 800 trench filling
TEOS/B(OMe)3/P(OMe)3 LPCVD 675 - 750 flow glass
TEOS/B(OMe)3/O=P(OMe)3 LPCVD 680 flow glass
diffusion
TEOS/AsCls APCVD 500 - 700
source
TEOS/As(OEt)3 LPCVD 700 - 730 trench doping
TEOS/O=As(OEt)3 LPCVD 700 - 730 trench doping

Precursors and deposition conditions for CVD of borosilicate glass (BSG), phososilicate glass
(PSG), borophosphosilicate glass (BPSG) and arsenosilicates (AsSG) thin films.

CVD from hydrides

Films of BSG, PSG, and BPSG may all be grown from SiHy, O» and B>Hg¢ and/or PHs, at 300 -
650 °C. For APCVD, the reactants are diluted with an inert gas such as nitrogen, and the
O,/hydride molar ratio is carefully controlled to maximize growth rate and dopant concentration
(values of 1 to 100 are used depending on the application). Ordinarily, the dopant concentration
for both BSG and PSG decreases with increased temperature. However, some reports indicate
an increase in boron content with increased temperature. Film growth of BPSG was found to
occur in two temperature regions. Deposition at low temperature (270 - 360 °C) occurred via a
surface reaction rate limiting growth (E, = 39 kcal/mol), while at higher temperature (350 - 450
°C), a mass-transport rate limited reaction region is observed (E, = 7.6 kcal/mol).

LPCVD of BSG and PSG is conducted at 450 - 550 °C with an O»:hydride ratio of 1:1.5.
Conversely, an Oy:hydride ratio of 1.5:1 provides the optimum growth conditions for BPSG
over the same temperature range. The phosphorous in PSG films was found to exist as a mixture
of P»Os and P»03, however, the latter can be minimized under the correct deposition conditions.
Some difficulties have been reported for the use of B>Hg due to its thermal instability.
Substitution of ByHg with BCl3 obviates this problem, although the resulting films are
invariably contaminated with 1 weight per cent chloride.

Arsenosilicate glass (AsSG) thin films are generally grown by APCVD using arsine (AsH3); the
use of which is being limited due to its high toxicity. However, arsine inhibits the gas phase
reactions between SiH, and Oy, such that film grown from SiH4/AsH3/O> show improved step
coverage at high deposition rates.

CVD from metal organic precursors

As with SiO, deposition, see above, there has been a trend towards the replacement of SiH, with
TEOS on account of its ability to produce highly conformal coatings. This is particularly
attractive with respect to trench filling. Furthermore, films of doped SiO» glasses have been
obtained using both APCVD and LPCVD (typically below 3 Torr), with a wide variety of
dopant elements including: boron, phosphorous, and arsenic, including antimony, tin, and zinc.

Boron-containing glasses are generally grown using either trimethylborate, B(OMe)s, or
triethylborate, B(OEt)3, although the multi-element source, tris(trimethylsilyl)borate,
B(OSiMe3)3, has been employed for both silicon and boron in BPSG thin film growth.
Similarly, whereas PH3 may be used as the phosphorous source, trimethylphosphite, P?COMe)s,
and trimethylphosphate, O=P(OMe)s, are preferred. Likewise, triethoxyarsine, As(OEt)3, and
triethylarsenate, O=As(OEt)3, have been employed for AsSG growth.

The co-reaction of TEOS with organoboron and organophosphorous compounds allows for
deposition at lower temperatures (500 - 650 °C) than for hydride growth of comparable rates.
However, LPCVD, using an all organometallic approach, requires P?(OMe)3 because the low
reactivity of O=P(OMe)3 prevents significant phosphorus incorporation. Although premature
decomposition of P}(OMe)3 occurs at 600 °C (leading to non-uniform growth), deposition at 550
°C results in high film uniformity at reasonable deposition rates.

Bibliography

W. Kern and V. S. Ban, in Thin Film Processes, Eds. J. L. Vossen, W. Kern, Academic
Press, New York (1978).

M. L. Hammod, Sold State Technol., 1980, 23, 104.

A. R. Barron and W. S. Rees, Jr., Adv. Mater. Optics Electron., 1993, 2, 271.

N. Goldsmith and W. Kern, RCA Rev., 1967, 28, 153.

C. Pavelescu, J. P. McVittie, C. Chang, K. C. Saraswat, and J. Y. Leong, Thin Solid Films,
1992, 217, 68.

J. D. Chapple-Sokol, C. J. Giunta, and R. G. Gordon, J. Electrochem. Soc., 1987, 136,
2993.

P. Gonzalez, D. Fernandez, J. Pou, E. Garcia, J. Serra, B. Leon, and M. Pérez-Amor, Thin
Solid Films, 1992, 218, 170.

E. L. Jordan, J. Electrochem. Soc., 1961, 108, 478.

K. Fujino, Y. Nishimoto, N. Tokumasu, and K. Maeda, J. Electrochem. Soc., 1990, 137,
2883.

R. A. Levy and K. Nassau, J. Electrochem. Soc., 1986, 133, 1417.

L. K. White, J. M. Shaw, W. A. Kurylo, and N. Miszkowski, J. Electrochem. Soc., 1990,
137, 1501.

Chemical Vapor Deposition of Alumina

Alumina

Alumina, Al)O3, exists as multiple crystalline forms, however, the two most important are
the a and y forms. a-Al,O3 (corundum) is stable at high temperatures and its structure
consists of a hexagonal close-packed array of oxide (O*”) ions with the Al°* ions
occupying octahedral interstices. In contrast, y-Al,O3 has a defect spinel structure, readily
takes up water and dissolves in acid. Despite the potential disadvantages of y-Al,O3 there
is a preference for its deposition on silicon substrates because of the two different lattice-
matching relationships of y-Al)O3 (100) on Si(100). These are shown as schematic
diagrams in [link]. A summary of CVD precursor systems for Al»O3 is given in [link].

i.

y-Al203

(a) (b)

Schematic diagram of the crystallographic relations of y-

Al,O3 on Si(100): (a) y-Al5O3 (100)||Si(100), and (b) y-

Al5O3 (100)||Si(110). Adapted from A. R. Barron, CVD
of Non-Metals, W. S. Rees, Jr., Ed. VCH, New York

(1996).
Aluminum Oxygen Carrier CVD Deposition
"i Comments
precursor source gas method temp. (°C)
AICls CO>/H> H> or APCVD 700 - 900 amorphous

No (700),

AlMe3

AlMe3

AlMe3

AlMe3

AlMe3

Al(O'Pr)3

Al(O'Pr)3

Al(O'Pr)3

Al(acac)3

Al(acac)3

Al(acac)3

O»

Op

N,O

air

O» and
H,O

Np> or
He

No

N> or

He

Np

Np

Np

Ar

APCVD

LPCVD

APCVD

LPCVD

PECVD

APCVD

LPCVD

LPCVD

APCVD

APCVD

LPCVD

350 - 380

375

100 - 660

950 - 1050

120 - 300

420 - 600

250 - 450

200 - 750

420 - 450

250 - 600

230 - 550

crystalline
(850 - 900)

dep. rate
highly
dependent
on gas-
phase
conc. Al
and O»

plasma-
enhanced,
10 W

lower
quality
than with
Oz

good
passivation
properties
of Si MOS
devices

plasma-
enhanced,

epitaxial
on Si

high C
content

significant
C content

growth rate
indep. of

H,O but
film
quality
dep. on
H,O

Precursors and deposition conditions for Al,O3 CVD.

CVD from halides

The initial use of CO>/Hp> as a hydrolysis source for the CVD of SiO> from SiCly, led to the
analogous deposition of Al»O3 from AICla, i.e.,
Equation:

H, + CO, > H,O + CO

Deposition in the temperature range 700 - 900 °C was found to yield films with optimum
dielectric properties, but films deposited below 700 °C contained significant chloride
impurities. It has been determined that H2O vapor, formed from Hy and COsy, acts as the
oxygen donor, and not the CO). The crystal form of the CVD-grown alumina films was
found to depend on the deposition temperature; films grown below 900 °C were y-Al)O3,
while those grown at 1200 °C were a-Al,O3, in accord with the known phase diagram for
this material.

CVD from trimethylaluminum (TMA)

Although trimethylaluminum, AlMe3 (TMA), reacts rapidly with water to yield Al)Os, the
reaction is highly exothermic (-1243 kJ/mol) and thus difficult to control. The oxygen
gettering properties of aluminum metal, however, can be employed in the controlled
MOCVD growth of Al»O3. The common deposition conditions employed for CVD of
Al,O3 from AlMe3 are similar to those used for aluminum-metal CVD, but with the
addition of an oxygen source, either O> or N50.

Films grown by APCVD using N>O are of inferior quality to those employing O>, due to
their exhibiting some optical absorption in the visible wavelength region. The growth of
high quality films using either oxygen source is highly dependent on the gas phase
concentrations of aluminum and “oxygen”. Further improvements in film quality are
observed with the use of a temperature gradient in the chambers deposition zone.

Attempts to lower the deposition temperature employing PECVD have been generally
successful. However, a detailed spectroscopic study showed that the use of NO as the
oxygen source resulted in significant carbon and hydrogen incorporation at low
temperatures (120 - 300 °C). The carbon and hydrogen contamination are lowered at high
deposition temperature, and completely removed by a post-deposition treatment under O>.
It was proposed that the carbon incorporated in the films is in the chemical form of Al-CH3
or Al-C(O)OH, while hydrogen exists as Al-OH moieties within the film.

Photo-assisted CVD of Al,O3 from AlMe3 has been reported to provide very high growth
rates (2000 A/min) and give films with electrical properties comparable to films deposited
using thermal or plasma techniques. Irradiation with a 248 nm (KrF) laser source allowed
for uniform deposition across a 3" wafer. However, use of 193 nm (ArF) irradiation
required dilution of the AlMe3 concentration to avoid non-uniform film growth.

CVD from alkoxides and B-diketonates

The pyrophoric nature of AlMe3 urged investigations into alternative precursors, in
particular those which already contain oxygen. Alternative precursors might also provide
possible routes to eliminate carbon contamination. Given the successful use of TEOS in
SiO, thin film growth, an analogous alkoxide precursor approach is logical. The first report
of Al,O3 films grown by CVD used an aluminum alkoxide precursors.

Aluminum tris-iso-propoxide, Al(O'Pr)3, is a commercially available inexpensive alkoxide
precursor compound. Deposition may be carried-out by either APCVD or LPCVD, using
oxygen as an additional oxidation source to ensure low carbon contamination. It is
adventitious to use LPCVD (10 Torr) growth to inhibit gas phase homogeneous reactions,
causing formation of a powdery deposit. The use of lower chamber pressures (3 Torr) and
N>O as the oxide source provided sufficient improvement in film quality to allow for
device fabrication.

The deposition of Al,O3 films from the pyrolysis of aluminum acetylacetonate, Al(acac)3
([link]a), has been widely investigated using both APCVD and LPCVD. The perceived
advantage of Al(acac)3 over other aluminum precursors includes lowered-toxicity, good
stability at room temperature, easy handling, high volatility at elevated temperatures, and
low cost. However, the quality of films was originally poor; carbon being the main
contaminant resulting from the thermolysis and incorporation of acetone and carbon
dioxide formed upon thermal decomposition ((link]).

(a) (b) (c)

Aluminum £-diketonate precursors.

acetone

200 250 300 350 400
Temperature of Pyrolysis (°C)

Gaseous decomposition products from the
pyrolysis of Al(acac)3 as a function of
pyrolysis temperature (Data from J. Von
Hoene, R. G. Charles, and W. M. Hickam, J.
Phys. Chem., 1958, 62, 1098).

Incomplete oxidation of the film may be readily solved by the addition of water vapor to
the carrier gas stream; pure carbon-free films being grown at temperatures as low as 230
°C. In fact, water vapor plays an important role in the film growth kinetics, film purity, and
the surface morphology of the grown films. While the growth rate is unaffected by the
addition of water vapor, its influence on the surface morphology is significant. Films
grown without water vapor on the Al,O3 surface is rough with particulates. In contrast,
films grown with water vapor are mirror smooth.

A systematic study of the kinetics of vaporization of Al(acac)3 along with fluorinated
aluminum f-diketonate complexes, Al(tfac)3 ([link]b) and Al(hfac)s ([link]c), has been
reported, and the saturation vapor pressures determined at 75 - 175 °C.

Aluminum silicates

The high dielectric constant, chemical stability and refractory character of aluminosilicates,
(Al,03),(SiO2)y, makes them useful as packaging materials in IC chip manufacture.
Mullite (3A1,03.2SiO>) prepared by sol-gel techniques, is often used as an encapsulant for
active devices and thin-film components. Amorphous alumina-silica films have also been
proposed as insulators in multilevel interconnections, since they do not suffer the
temperature instability of alumina films retain the desirable insulating characteristics.
Under certain conditions of growth and fabrication, silica may crystallize, thereby allowing
diffusion of oxygen and impurities along grain boundaries to the silicon substrate
underneath. Such unwanted reactions are catastrophic to the electronic properties of the
device. The retention of amorphous structure over a larger temperature range of silicon rich
alumina-silica films offers a possible solution to this deleterious diffusion.

Thin films of mixed metal oxides are usually obtained from a mixture of two different
kinds of alkoxide precursors. However, this method suffers from problems with
stoichiometry control since extensive efforts must be made to control the vapor phase
concentration of two precursors with often dissimilar vapor pressures. Also of import here
is the near impossible task of matching rates of hydrolysis/oxidation to give "pure", non-
phase segregated films, i.e., those having a homogeneous composition and structure. In an
effort to solve these problems, research effort has been aimed at single-source precursors,
i.e., those containing both aluminum and silicon.

The first study of single-source precursors for (Al203),(SiO2), films employed the mono-
siloxide complex Al(O'Pr):(OSiMes) ([link]a). However, it was found that except for
deposition at very high temperatures (> 900 °C) the deposited films this mono-siloxide
compound were aluminum-rich (AI/Si = 1.3 - 2.1) and thus showed thermal instability in
the insulating properties caused by crystallization in the films. It would appear that in order
for silicon-rich alumina-silica films to be grown more siloxane substituents are required,
e.g., the tris-siloxy aluminum complex [Al(OSiEt3)3]» ([link]b).

SiEt, SiEt;

'PrOn., Ven 1 \O'Pr EnSiOv..,/” Ng

ae a Sas pnOSiEt,
iPpror WA ~oPr Sore NZ Osi
SiEt, SiEt,
(a) (b)

Precursors for aluminum silicate thin films.

The AI/Si ratio of thin films growth by APCVD using [AI(OSiEt3)3]5 at 420 - 550 °C, was
found to be dependent on the deposition temperature and the carrier gas composition
(O>/Ar). This temperature and oxygen-dependent variation in the film composition
suggests that two competing precursor decomposition pathways are present.

1. Deposition in the absence of Oy, is similar to that observed for the decomposition of
Al(O'Pr),(OSiMe3) under Np, and would imply that the film composition is
determined by the temperature-dependent tendencies of the Al-O-Si bonds to cleave.

2. The temperature-independent oxidative decomposition of the precursor. While it is
possible to prepare films richer in Si using [Al(OSiEt3)3]> rather than
Al(O'Pr)2(OSiMes), the Al:Si ratio is unfortunately not easily controlled simply by the
number of siloxy ligands per aluminum in the precursor.

Films grown from the single-source precursor Al(O'Pr)»(OSiMe;) crystallize to kyanite,
Al)SiOs, whereas those grown from [Al(OSiEt3)3]> remained amorphous even after
annealing.

Bibliography

e A.W. Apblett, L. K. Cheatham, and A. R. Barron, J. Mater. Chem., 1991, 1 ,143.

e K. M. Gustin and R. G. Gordon, J. Electronic Mater., 1988, 17, 509.

e C. Landry, L. K. Cheatham, A. N. MacInnes, and A. R. Barron, Adv. Mater. Optics
Electron., 1992, 1, 3.

e Y. Nakaido and S. Toyoshima, J. Electrochem. Soc., 1968, 115, 1094.

e T. Maruyama and T. Nakai, Appl. Phys. Lett., 1991, 58, 2079.

e K. Sawada, M. Ishida, T. Nakamura, and N. Ohtake, Appl. Phys. Lett., 1988, 52, 1673.

e J. Von Hoene, R. G. Charles, and W. M. Hickam, J. Phys. Chem., 1958, 62, 1098.

Introduction to Nitride Chemical Vapor Deposition

The refractory nature and high dielectric properties of many nitrides make
them attractive for chemical and electronic passivation. As a consequence
silicon nitride has become the standard within the semiconductor industry,
as both an encapsulation layer and as an etch mask.

In a similar manner to oxide growth by chemical vapor deposition (CVD),
two sources are generally required for binary nitride CVD: the element of
choice and a nitrogen source. However, unlike the CVD of oxides,
elemental nitrogen (N>) is not reactive, even at elevated temperatures,
thereby requiring plasma enhancement. Even with plasma enhanced CVD
(PECVD), N> does not yield high quality films. As a substitute for N>,
ammonia (NH3) has found general acceptance as a suitable nitrogen source.
It is a gas, readily purified and cheap, however, it is of low reactivity at low
temperatures. PECVD has therefore found favor for low temperature NH3-
based precursor systems.

Recent attempts to lower deposition temperatures have included the use of
more reactive sources (e.g., Hy NNH>) and precursors containing nitrogen as
a coordinated ligand. Probably the most important discovery with respect to
nitride deposition is the use of a transamination reaction between amido
compounds and ammonia ((link]).

Equation:

M—NR, + NH; —————» M—NH, + HNR,

Bibliography

e D. M. Hoffman, Polyhedron, 1994, 13, 1169.

Chemical Vapor Deposition of Silicon Nitride and Oxynitride

Introduction

Stoichiometric silicon nitride (SizN,4) is used for chemical passivation and encapsulation of
silicon bipolar and metal oxide semiconductor (MOS) devices, because of its extremely good
barrier properties for water and sodium ion diffusion. Water causes device metallization to
corrode, and sodium causes devices to become electrically unstable. Silicon nitride is also used
as a mask for the selective oxidation of silicon, and as a strong dielectric in MNOS (metal-
nitride-oxide-silicon) structures.

The use of ion implantation for the formation of active layers in GaAs MESFET devices ([link])
allow for control of the active layer thickness and doping density. Since implantation causes
structural disorder, the crystal lattice of the GaAs must be subjected to a post implantation rapid
thermal anneal step to repair the damage and to activate the implanted species. The required
annealing temperature (> 800 °C) is higher than the temperature at which GaAs decomposes.
Silicon nitride encapsulation is used to prevent such dissociation. Silicon nitride is also used for
the final encapsulation of GaAs MESFET devices ((link]).

Source Gate Drain
contact contact contact

Contact metal

Gate layer
Buffer layer

Schematic diagrams of a GaAs metal-semiconductor
field effect transistor (MESFET). Adapted from A. R.
Barron, in CVD of Nonmetals, Ed. W. S. Rees, Jr., Wiley,
NY (1996).

The deposition of Si3N, is a broadly practiced industrial process using either grown by low
pressure CVD (LPCVD) or plasma enhanced CVD (PECVD) with comparable properties for the
grown films ([Llink]).

Deposition LPCVD PECVD

Growth temperature (°C)

Composition

Si/N ratio

Atom% H

Dielectric constant
Refractive index
Resistivity (Q.cm)

Band gap (eV)

Silicon
precursor

SiH,

SiH,

SiH,

SiCl,H>

SivCle
Et)SiH>

RSi(N3)3 (R

Nitrogen
source

Carrier

gas

700 - 800
Si3N4(H)

0.75

CVD
method

APCVD

PECVD

PECVD

LPCVD

LPCVD
LPCVD

LPCVD

250 - 350

SiN,Hy

0.8 - 1.2

20-25

6-9

1.8-2.5

106 - 1015

4-5

A summary of some typical CVD systems for silicon nitride is given in [Link].

Deposition
temp. (°C)

70 - 900

20 - 600

70 - 300

700 - 900

450 - 850
650 - 725

450 - 600

Summary of the properties of silicon nitride grown in typical commercial systems.

One of the disadvantages of Si3N, is its high dielectric constant that may limit device speed at
higher operating frequencies. It is hoped that silicon oxynitride (SiON) films will exhibit the best
properties of Si3N4 and SiO», namely the passivation and mechanical properties of Si3N, and the
low dielectric constant and low stress of SiO>.

Comment

Commercial
process

Porous films

Commercial
process

C impurities

Danger —

= Et, ‘Bu) precursor

explosive
MeSiH(NH), —- NH3/H, | APCVD — 600-800 ees C
Si(NMep)4. - He APCVD 600 - 750 Significant C
nHp content
a NH3 He APCVD 600-750 NOE ate
ny contamination

Precursors and deposition conditions for SizN4 CVD.

CVD of silicon nitride from hydrides and chlorides

The first commercial growth of silicon nitride was by the reaction of SiH, and NH3 by either
atmospheric pressure CVD (APCVD) or PECVD. Film growth using APCVD is slower and
requires higher temperatures and so it has been generally supplanted by plasma growth, however,
film quality for APCVD is higher due to the lower hydrogen content. While thermally grown
films are close to stoichiometric, PECVD films have a composition in which the S/N ratio is
observed to vary from 0.7 - 1.1. The non-stoichiometric nature of PECVD films is explained by
the incorporation of significant hydrogen in the films (10 - 30%). PECVD of SiN, using SiH,/N>
leads to electronically leaky films due to the porous nature of the films, however, if an electron
cyclotron resonance (ECR) plasma is employed, SiNx films of high quality may be deposited on
ambient temperature substrates.

The more recent commercial methods for silicon nitride deposition involves LPCVD using
SiCl)H) as the silicon source in combination with NH3 at 700 - 900 °C. The reduced pressure of
LPCVD has the advantages of high purity, low hydrogen content, stoichiometric films, with a
high degree of uniformity, and a high wafer throughput. It is for these reasons that LPCVD is
now the method of choice in commercial systems. A large excess of NH3 is therefore used in
commercial systems to obtain stoichiometric films. Silicon nitride has also been prepared from
SiCl4/NHy, SiBr4/NH3, and, more recently, SipCl¢/NH3.

Silicon oxynitride (SiON) may be prepared by the use of any of the precursors used for silicon
nitride with the addition of either N»O or NO as an oxygen source. The composition and
properties of the SiO,N, films may be varied from SiOQ>-like to SigN,-like by the variation of the
reactant flow rates.

SiCl)H> gas plumbing to a LPCVD reactor must be thermally insulated to prevent condensation
that would otherwise lead to hazy deposits on the film. The volatile by-products from CVD
produce NH,Cl at the exhaust of the reaction tube, and in the plumbing and pumping system. It
would be desirable, therefore, to find an alternative, chlorine-free silicon source with none of the
toxicity or pyrophoricity problems associated with SiHy,. It is for this reason that organosilicon
compounds have been investigated.

CVD from organosilicon precursors

Diethylsilane, EtpSiHz, has shown promise as a replacement for SiH, in the low temperature
LPCVD of SiOs, and has been investigated as a source for SiN, and SiON, films. Deposition by
LPCVD in the presence of NH3 produces SiN, films, in which the carbon contamination (4 -
9%) depends on the partial pressure of the EtySiH». The presence of carbon raises the refractive
index (2.025 - 2.28) with respect to traditional LPCVD films (2.01). Mixtures of EtpSiH2, NH3,
and NO deposit SiO,N, films where the composition is controlled by the NH3:N,O ratio.

CVD from silicon-nitrogen compounds

The incorporation of carbon into silicon nitride films is a persistent problem of organosilicon
precursors. Several studies have been aimed at developing single source precursors containing a
Si-N bond rather than Si-C bonds. Polyazidosilanes, R,Si(N3)4_,, are low in carbon and
hydrogen, reasonably volatile, and contain highly activated nitrogen, however, they represent a
significant explosive hazard: they are explosive with an equivalent force to TNT. Films deposited
using EtSi(N3)3 and (tBu)Si(N3)3 showed promise, despite the observation of oxygen and
carbon. Pyrolytic studies on the azide precursors suggest that the primary decomposition step is
the loss of dinitrogen, which is followed by migration of the alkyl onto the remaining nitrogen,
[link]. The fact that neither the addition of NH3 or H> influence the film deposition rate suggest
that the intramolecular nitride formation process is fast, relative to reaction with NH3, or

hydrogenation.
Equation:
R R Ng
A /
N,—N—Si —— 3 NS, -- = 3& RNS
i VN
3 3 N.
3 Ng 7

Carbon incorporation is also observed for the APCVD deposition from Si(NMe>),H4_, (n = 2 -
4). However, using the Hoffman transamination reaction, deposition in the presence of NH3
completely removed carbon incorporation into the stoichiometric Si3N, film. From FTIR data,
the hydrogen content was estimated to be 8 - 10 atom percent. While the Si(NMe>),H4_,/NH3
system does not provide substantially lower temperatures than APCVD using SiH,/NH3 growth
rates are significantly higher. Unlike the azide precursors, Si(NMe>),H4_, are easier to handle
than either SiH, or SiCljH>.

Bibliography

e J.C. Barbour, H. J. Stein, O. A. Popov, M. Yoder, and C.A. Outten, J. Vac. Sci. Technol. A.,
1991, 9, 480.

e J. A. Higgens, R. L. Kuvas, F. H. Eisen, and D. R. Chen, IEEE Trans. Electron. Devices,
1978, 25, 587.

e D.M. Hoffman, Polyhedron, 1994, 13, 1169.

e W. Kellner, H. Kniepkamp, D. Repow, M. Heinzel, and H. Boroleka, Solid State Electron.,
1977, 20, 459.

e T. Makino, J. Electrochem. Soc., 1983, 130, 450.

e C. T. Naber and G. C. Lockwood, in Semiconductor Silicon, Eds. H. R. Huff and R. R.
Burgess. The Electrochemical Society, Softbound Proceedings Series, Princeton, NJ (1973).

e J. E. Schoenholtz, D. W. Hess, Thin Solid Films, 1987, 148, 285.

Chemical Vapor Deposition of Aluminum Nitride

Introduction

Aluminum nitride (AIN) has potential for significant applications in microelectronic and
optical devices. It has a large direct bandgap (Eg qj = 6.28 eV), extremely high melting point
(3000 °C), high thermal conductivity (2.6 W/cm.K), and a large dielectric constant (€ = 9.14).
In present commercial microelectronic devices, AIN is used most often as a packaging
material, allowing for the construction of complex packages with many signal, ground, power,
bonding, and sealing layers. Aluminum nitride is especially useful for high power applications
due to its enhanced thermal conductivity. Chemical vapor deposition (CVD) grown thin films
of AIN have been centered upon its use as a high gate-insulation layer for MIS devices, and a
dielectric in high-performance capacitors. One additional property of AIN that makes it a
promising insulating material for both Si and GaAs devices is that its thermal expansion
coefficient is almost identical to both of these semiconductors.

The lack of a suitably volatile homoleptic hydride for aluminum (A1Hs is an involatile
polymeric species) led to the application of aluminum halides and organometallic compounds
as precursors. A summary of selected precursor combinations is given in [link].

Aluminum Nitrogen Carrier CVD Deposition

precursor source gas method temp. (°C) Sour
AICl3(NH3) - Np LPCVD 700 - 1400 ees
present
AIBrs3 NH3 N> APCVD 400 - 900 Br present
AIBr3 N5 N5 LPCVD 520-560 oe
growth
AlMe3 NH; H> LPCVD 1200
AlMe3 NH3 He APCVD 350 - 400
N-H and
ae AIN-N
AlMe3 cracked H,/He APCVD 310 - 460
bonds
NH;
detected

AlMe, ‘BuNH, Hp APCVD 400 - 600 high C

or content,
*PrNH> low N

AlMe3 Me3SiN3 Hy APCVD 300 - 450 pe
poor film
quality,
high C
content

[R2Al(NH>)]3

ean H> LPCVD 400 - 800

unreacted
[RoAIN3]3 (R - LPCVD 400 - 500 precursor
= Me, Et) present on
film

amorphous
100 - 200
dO
crystalline
300 - 500
°C

Al(NMe;)3 NH; He APCVD 100-500

Precursors and deposition conditions for AIN CVD.

CVD from halides

The observation that AIN powder may be produced upon the thermal decomposition of the
AICl3(NH3) complex, prompted initial studies on the use of AlCl3/NH3 for the CVD of AIN
films. Initially, the low volatility of AlCl; (a polymeric chain structure) required that the
AICl3(NH3) complex to be used as a single precursor. Low pressure CVD (LPCVD) at 5 -10
Torr resulted in deposition of AIN films, although films deposited below 1000 °C were
contaminated with NH,Cl, and all the films contained chlorine. Films with reasonable
electrical properties were prepared by the use of the more volatile tris-ammonia complex,
AICIl3(NH3)3. The dielectric constant for films grown at 800 - 1000 °C (11.5) is higher than
bulk AIN (9.14) and also than that of the films grown at 1100 °C (8.1). All the films were
polycrystalline with the grain size increasing with increasing deposition temperatures and
preferred orientation was observed only for the films grown below 1000 °C.

Aluminum bromide is a dimeric volatile compound, [BrjAl(p-Br)]>, and is more attractive as a
CVD source, than AICl3. Deposition of AIN films can be accomplished using AlBr3 and NH3
in an APCVD system with Hp as the carrier gas. The mechanism of film growth has been
proposed ([link]).

AIBr(g) + NH3(g)

surface adsorption
homogeneous surface reaction

reaction AIBr,.NH3(surface) ——————————_»_ AIN

decomposition
of intermediate film
surface adsorption compounds

AIBr,(NH3 )(g)
Mechanism of APCVD film growth of AIN using AIBrs
and NH3.

Due to the high temperatures required (750 °C) for good quality AIN film growth from AIBrs,
PECVD was investigated. Using an AlBr3-H»-N> gas mixture and a 2450 MHz microwave
(100 - 1000 W) plasma source, AIN films were grown. The maximum deposition rate occurred
with an N>/AIBr3 ratio of ca. 20 and a substrate temperature ca. 600 °C.

CVD from aluminum alkyls

Based upon the successful metal organic CVD (MOCVD) growth of AlGaAs using the alkyl
derivatives, AIR3, it was logical to extend MOCVD to aluminum nitride. Initial studies were
performed using AlMe3 and NH3 with H) carrier gas. While these films are generally of high
quality, the temperature of deposition is incompatible with semiconductor processing (being
above both the melting point of most metallization alloys and the temperature at which dopant
migration becomes deleterious). Lower temperatures (as low as 350 °C) were explored,
however significant pre-reaction was observed between AlMe3 and NH3; causing depletion of
the reactants in the deposition zone, reducing the growth rate and leading to non-uniform
deposits. Two routes have been investigated by which this problem can be circumvented.

PECVD successfully lowers the deposition temperature, although, degradation of the substrate
surface by ion bombardment is a significant drawback. Given that it is the ammonia
decomposition that represents the highest energy process, pre-cracking should lower the
overall deposition temperature. This is indeed observed for the AlMe3/NH3-based AIN system
where growth is achieved as low as 584 °C if the NHs3 is catalytically cracked over a heated
tungsten filament (1747 °C). In fact, with catalytic pre-cracking, deposition rates were
observed to be an order of magnitude greater than for PECVD at the same temperatures,
resulting in films that were crystalline with columnar growth. For this approach to low-
temperature MOCVD growth of AIN the only major drawback is the presence of residual N-H
and AIN-N groups detected by FT-IR.

Chemical solutions to the high stability of NH3 have primarily centered upon the use of
alternative nitrogen sources. The use of the volatile nitrogen source hydrazine (N>Hy), has
allowed for the growth of AIN at temperatures as low as 220 °C, however, hydrazine is
extremely toxic and highly unstable, restricting its commercial application. Primary amines,
such as ‘BuNH) or 'PrNHp, allow for deposition at modest temperatures (400 - 600 °C). The

high carbon incorporation, as high as 17% precludes their adoption. A similar problem is
observed with the use of trimethylsilylazide, Me3SiN3. The presence of carbon contamination
in the deposition of Al films and AlGaAs epitaxial layers has been attributed to the use of
AlMe3. Therefore attempts have been made to use alternative aluminum precursors.

Interest in the mechanism of nucleation and atomic layer growth of AIN has prompted several
mechanistic studies of the formation of Al-N bonds on the growth surface. All the studies
concurred that the mechanism involves a step-wise reaction where the amide (-NH>-) groups
form covalent bonds to aluminum irrespective of substrate. A schematic representation of the
process is shown in [link].

H3C. NH. NH.

ee ee a ab
4: 4: 4: fc

O oO oO Oo —_—_——_
ie loetiog ao ies eee
Si Si Si Si i j i i

cata a

AL 2 AL ¢s Al
VS ONS
2
}): ‘Al -

= £3 + AI(CH3)3

fodody Aa fo fe 9
CELLET ZECcAZZ--=- ZZ

ga

A schematic representation of the proposed step-
wise reaction involving the formation of amide (-
NH)-) groups covalently bound to aluminum
during the MOCVD growth of AIN using
AlMe;3/NHs3. (Adapted from M. E. Bartram, T. A.
Michalske, J. W. Rogers, Jr., and R. T. Paine,
Chem. Mater., 1993, 5, 1424).

CVD from aluminum amide and related compounds

The reaction between aluminum alkyls and amines ([link]), as well as the formation of AIN
powders from the pyrolysis of AlMe3(NHs3) ([link]), lead to the misguided concept that the
route to high-purity AIN would be through the so-called single source precursor route.
Equation:

AIR, + HNR', —4> 1/[R,AI(NR')], + RH

Equation:

AlMe,(NH;) > AIN + 3 MeH

The trimeric dimethylaluminum amide, [Me,Al(NH>)]3 ([link]a), was originally used as a
single source precursor for growth of AIN under LPCVD conditions using a hot walled
reactor, although subsequent deposition was also demonstrated in a cold walled system. Film
quality was never demonstrated for electronic applications, but the films showed promise as
fiber coatings for composites. The concept of using a trimeric single source precursor for AIN
was derived from the observation of Al3N3 cycles as the smallest structural fragment in
wurtzite AIN. However, detailed mechanistic studies indicate that under gas phase thermolysis
the trimeric precursor [Me Al(NH2)]3 is in equilibrium with (or decomposes to) dimeric
({link]b) and monomeric ({link]c) compounds. Furthermore, nitrogen-poor species ({link]d)
were also observed by TOF-mass spectrometry.

Me Me
cf
a . H
Al Al.
Me™" \'Me
Me* re Me
HH
HH (a) Me
Ri M H I H
e, 1 “N
Mer / \ wwwMe \ HA N~
mo AK AIS Me Does | |
\ Me H VE i
H H Me Me Me Me

(b) (c) (d)

The trimeric dimethylaluminum amide (a)
used as a single source precursor for growth
of AIN, and the decomposition products (b -

d) observed by TOF-mass spectrometry.

Following the early reports of single source precursor routes, a wide range of compounds have
been investigated, including [Al(NR>)3]2, [HAI(NR»)212 (R = Me, Et), and [Me,AIN(Pr)9]p,
all of which gave AIN, but none of these precursors give films of superior quality comparable
to that obtained from traditional CVD. In particular, the films contained significant carbon
contamination, prompting further investigations into the efficacy of, N-C bond free,
dialkylaluminum azides, [R>AI(N3)]3, as LPCVD precursors.

While aluminum tris-amides, Al(NR>)3 were shown to give carbon-contaminated films,
APCVD carried-out with NH3 as the carrier gas results in carbon-free AIN film growth as low
as 100 °C. The reason for the deposition of high quality films at such low temperatures resides

with the Hoffman transamination reaction between the primary amido unit and ammonia. The
crystallinity, bandgap and refractive index for the AIN grown by APCVD using [Al(NMe3)3].
and NHg3 are dependent on the deposition temperature. Films grown at 100 - 200 °C are
amorphous and have a low bandgap and low refractive index. Above 300 °C, the films are
crystalline, and have a refractive index close to that of bulk AIN (1.99 - 2.02), with a bandgap
(< 5.77 eV) approaching the values reported for polycrystalline AIN (5.8 - 5.9 eV).

Bibliography

e J. L. Dupuie and E. Gulari, J. Vac. Sci. Technol. A, 1992, 10, 18.

e D.M. Hoffman, Polyhedron, 1994, 13, 1169.

e L. V. Interrante, W. Lee, M. McConnell, N. Lewis, and E. Hall, J. Electrochem. Soc.,
1989, 136, 472.

e H. M. Manasevit, F. M. Erdmann, and W. I. Simpson, J. Electrochem. Soc., 1971, 118,
1864.

e Y. Pauleau, A. Bouteville, J.J. Hantzpergue, J. C Remy, and A. Cachard, J. Electrochem.
Soc.,. 1980, 127 1532,

e Y. Someno, M. Sasaki, and T. Hirai, Jpn. J. Appl. Phys., 1990, 29, L358.

Metal Organic Chemical Vapor Deposition of Calcium Fluoride

The chemical vapor deposition (CVD) of metal fluorides has been much
less studied than that of oxides, pnictides, or chalgogenides. As may be
expected where a volatile fluoride precursor is available then suitable films
may be grown. For example, Group 5 (V, Nb, Ta), 6 (Mo, W), and 7 (Re)
transition metals are readily deposited from fluoride-hydrogen mixtures.
While the use of fluorine is discouraged on safety grounds, many of the
fluorinated alkoxide or B-diketonate ligands employed for metal oxide
metal organic chemical vapor deposition (MOCVD) are predisposed to
depositing metal fluorides. The use of fluorine substituted derivatives is
because they are often more volatile than their hydrocarbon analogs, and
therefore readily used for both atmospheric and low pressure CVD. To
minimize the unwanted formation of metal fluorides, water vapor is
incorporated in the gas stream, and it is common to perform post-deposition
hydrolytic anneals. However, there exist a number of applications where
fluorides are required. For example, the highly insulating nature of CaF
and SrF> has prompted investigations into their use as a gate insulator in
GaAs-based metal insulator semiconductor field effect transistor (MISFET)
devices. It should be noted that while CaF) is a good insulator, the
CaF>/GaAs interface has a high interface trap density, requiring a
passivation buffer layer to be deposited on GaAs prior to CaF» growth.

One of the difficulties with the use of CaF, (and SrF>) on GaAs is the lattice
mismatch ([link]), but this may be minimized by the use of solid solutions
between CaF -SrF>. The composition Cag 44Sro.5¢F 2 is almost perfectly
lattice-matched to GaAs. Unfortunately, the thermal expansion coefficient
differences between GaAs and CaF>-SrF> produce strains at the
film/substrate interface under high temperature growth conditions. The
solution to this latter problem lies in the low temperature deposition of
CaF>-SrF) by CVD.

Compound Lattice constant (A)

CaF, 5.46

SrF> 5.86
BaF»> 6.20
GaAs 5.6532

Lattice parameters of Group 2 (II) fluorides in comparison with GaAs.

Polycrystalline CaF may be grown by the pyrolytic decomposition of
Ca(CsMes)> ([link]a) in either SiF, or NF3. Deposition at 150 °C results in
polycrystalline films with high levels of carbon (18%) and oxygen (7%)
impurities limiting the films usefulness in electronic applications. However,
significantly higher purity films may be grown at 100 °C using the photo-
assisted decomposition of Ca(hfac), ({link]b). These films were deposited at
30 A/min and showed a high degree of crystallographic preferred
orientation.

CF;
e jf
Me. O— \
Ca Me Cc. +) CH
Me O— q
M
2 2, CF;
2
(a) (b)

CaF», MOCVD precursors.

The mechanism enabeling fluoride transfer to the metal (from the carbon of
fluorinated alkoxide ligands) has been investigated. MOCVD employing
[Na(OR,)]4 and Zr(OR¢)4 [ORs = OCH(CF3)5 and OCMe3_,(CF3),, n = 1 -
3] gives NaF and ZrF, films, respectively, with volatile fluorocarbon side-
products. Analysis of the organic side-products indicated that

decomposition occurs by transfer of fluorine to the metal in conjunction
with a 1,2-migration of a residual group on the alkoxide, to form a ketone
({link]). The migration is increasingly facile in the order CF3 << CH3 < H.
The initial M-F bond formation has been proposed to be as a consequence
of the close MF agostic interactions observed for some fluoroalkoxide and
fluoro-B-diketonates.

F

ee me i
KEM eC ON 2 aa —» [(F,C)Me,CO};Zr—-F +
oO y
{- / me CF,Me
Mee

Zr Fy

Proposed mechanism for the decomposition of
fluorinated alkoxide compounds. (Adapted from J.
A. Samuels, W. -C. Chiang, C. -P. Yu, E. Apen, D.

C. Smith, D. V. Baxter, K. G. Caulton, Chem.
Mater., 1994, 6, 1684).

Bibliography

e A. R. Barron, in CVD of Nonmetals, W. S. Rees, Jr. (ed), Wiley, New
York (1996).

e B.D. Fahlman and A. R. Barron, Adv. Mater. Opt. Electron., 2000, 10,

273:

H. Heral. L. Bernard, A. Rocher, C. Fontaine, A. Munoz-Jague, J.

Appl. Phys., 1987, 61, 2410.

e J. A. Samuels, W. -C. Chiang, C. -P. Yu, E. Apen, D. C. Smith, D. V.
Baxter, K. G. Caulton, Chem. Mater., 1994, 6, 1684.

e W. Vere, K. J. Mackey, D. C. Rodway, P. C. Smith, D. M Frigo, D. C.
Bradley, Angew. Chem. Int. Ed. Engl. Adv. Mater., 1989, 28, 1581.

Precursors for Chemical Vapor Deposition of Copper

Note: This module was developed as part of the Rice University course CHEM-
496: Chemistry of Electronic Materials. This module was prepared with the
assistance of Wei Zhao.

Introduction

Chemical vapor deposition (CVD) is a process for depositing solid elements and
compounds by reactions of gas-phase molecular precursors. Deposition of a
majority of the solid elements and a large and ever-growing number of
compounds is possible by CVD.

Most metallization for microelectronics today is performed by the physical
vapor deposition (PVD) processes of evaporation and sputtering, which are
often conceptually and experimentally more straightforward than CVD.
However, the increasing importance of CVD is due to a large degree to the
advantages that it holds over physical vapor deposition. Foremost among these
are the advantages of conformal coverage and selectivity. Sputtering and
evaporation are by their nature line-of-sight deposition processes in which the
substrate to be coated must be placed directly in front of the PVD source. In
contrast, CVD allows any substrate to be coated that is in a region of sufficient
precursor partial pressure. This allows the uniform coating of several substrate
wafers at once, of both sides of a substrate wafer, or of a substrate of large size
and/or complex shape. The PVD techniques clearly will also deposit metal on
any surface that is in line of sight. On the other hand, it is possible to deposit
selectively on some substrate materials in the presence of others using CVD,
because the deposition is controlled by the surface chemistry of the
precursor/substrate pair. Thus, it may be possible, for example, to synthesize a
CVD precursor that under certain conditions will deposit on metals but not on
an insulating material such as SiO», and to exploit this selectivity, for example,
in the fabrication of a very large-scale integrated (VLSI) circuit. It should also
be pointed out that, unlike some PVD applications, CVD does not cause
radiation damage of the substrate.

Since the 1960s, there has been considerable interest in the application of metal
CVD for thin-film deposition for metallization of integrated circuits. Research
on the thermal CVD of copper is motivated by the fact that copper has physical
properties that may make it superior to either tungsten or aluminum in certain
microelectronics applications. The resistivity of copper (1.67 mW.cm) is much
lower than that of tungsten (5.6 mW.cm) and significantly lower than that of
aluminum (2.7 mW.cm). This immediately suggests that copper could be a
superior material for making metal interconnects, especially in devices where
relatively long interconnects are required. The electromigration resistance of
copper is higher than that of aluminum by four orders of magnitude. Copper has
increased resistance to stress-induced voidage due to its higher melting point
versus aluminum. There are also reported advantages for copper related device
performance such as greater speed and reduced cross talk and smaller RC time
constants. On the whole, the combination of superior resistivity and
intermediate reliability properties makes copper a promising material for many
applications, provide that suitable CVD processes can be devised.

Applications of metal CVD

There are a number of potential microelectronic applications for metal CVD,
including gate metallization (deposit on semiconductor), contact metallization
(deposit on semiconductor), diffusion barrier metallization (deposit on
semiconductor), interconnect metallization (deposit on insulator and conductor
or semiconductor). Most of the relevant features of metal CVD are found in the
interconnect and via fill applications, which we briefly describe here. There are
basically two types of metal CVD processes that may occur:

1. Blanket or nonselective deposition, in which deposition proceeds
uniformly over a variety of surfaces.

2. Selective deposition in which deposition only occurs on certain types of
surfaces (usually semiconductors or conductors, but not insulators).

A primary application of blanket metal CVD is for interconnects. The conformal
nature of the CVD process is one of the key advantages of CVD over PVD and
is a driving force for its research and development. The degree of conformality
is usually described as the “step coverage”, which is normally defined as the
ratio of the deposit thickness on the step sidewall to the deposit thickness on the
top surface. Another application for blanket metal CVD is via hole filling to

planarize each level for subsequent processing, This is achieved by depositing a
conformal film and etching back to the insulator surface, leaving the metal
“plug” intact. Another unique aspect of CVD is its potential to deposit films
selectively, which would eliminate several processing steps required to perform
the same task. The primary application for selective metal CVD would be for
via hole filling. Ideally, deposition only occurs on exposed conductor or
semiconductor surfaces, so filling of the via hole is achieved in a single step.

Copper CVD

The chemical vapor deposition of copper originally suffered from a lack of
readily available copper compounds with the requisite properties to serve as
CVD precursors. The successful development of a technologically useful copper
CVD process requires first and foremost the design and synthesis of a copper
precursor which is volatile, i.e., possesses an appreciable vapor pressure and
vaporization rate to allow ease in transportation to the reaction zone and
deposition at high growth rates. Its decomposition mechanism(s) should
preferably be straightforward and lead to the formation of pure copper and
volatile by-products that are nonreactive and can be cleanly removed from the
reaction zone to prevent film, substrate, and reactor contamination. Gaseous or
liquid sources are preferred to solid sources to avoid undesirable variations in
vaporization rates because of surface-area changes during evaporation of solid
sources and to permit high levels of reproducibility and control in source
delivery. Other desirable features in precursor selection include chemical and
thermal stability to allow extended shelf life and ease in transport and handling,
relative safety to minimize the industrial and environmental impact of
processing and disposal, and low synthesis and production costs to ensure an
economically viable process.

Several classes of inorganic and metalorganic sources have been explored as
copper sources. Inorganic precursors for copper CVD used hydrogen reduction
of copper halide sources of the type CuX or CuX», where X is chlorine (Cl) or
fluorine (F):

BCU tg 32 Gu 2 FLX

CuX) + Hy - Cu+2 HX

The volatility of copper halides is low, the reactions involved require
prohibitively high temperatures (400 - 1200 °C), lead to the production of
corrosive by-products such as hydrochloric and hydrofluoric acids (HCl and
HF), and produce deposits with large concentrations of halide contaminants.
Meanwhile, the exploration of metalorganic chemistries has involved various
copper(II) and copper(I) source precursors, with significant advantages over
inorganic precursors.

From Cu(II) precursors

Volatile Cu(I]) compounds

Copper was known to form very few stable, volatile alkyl or carbonyl
compounds. This was thought to eliminate the two major classes of compounds
used in most existing processes for CVD of metals or compound
semiconductors. Copper halides have been used for chemical vapor transport
growth of Cu-containing semiconductor crystals. But the evaporation
temperatures needed for copper halides are much higher than those needed for
metal-organic compounds. Film purity and resistivity were also a problem,
possibly reflecting the high reactivity of Si substrates with metal halides.

Cu(II) compounds that have been studied as CVD precursors are listed in [link].
The structural formulas of these compounds are shown in [link] along with the
ligand abbreviations in [link]. Each compound contains a central Cu(II) atom
bonded to two singly charged f-diketonate or B-ketoiminate ligands. Most of
them are stable, easy to synthesize, transport and handle.

Evaporation Deposition Carrier | Reactor
Suepoun temp. (°C) temp. (°C) as pressure
‘i a (Torr)
Cu(acac), 180 - 200 225 - 250 H/Ar 760

Cu(hfac), 80 - 95 250 - 300 Hy 760

Cu(tfac), 135 - 160 250 - 300 H> 760

Cu(dpm), 100 400 none <10-4

Cu(ppm)> 100 400 none <0.3

Cu(fod), ‘ 300 - 400 Hp 10™ -
760

Cu(acim)> 287 400 Hp 730

oer 85 - 105 270 - 350 He 10-70

2
Cu(acen)> 204 450 H> 730

Studies of Cu CVD using Cu(II) compound. Adapted from T. Kodas and M.
Hampden-Smith, The Chemistry of Metal CVD, VCH Publishers Inc., New York,
NY (1994).

HQ 6 9

(a) (b) (c)

Structures of Cu(II) compounds studied as CVD precursors.

Ligand Structural

abbreviation Pe 7 type
acac CH3 CH3 a
hfac CF3 CF3 a
tfac CH3 CF3 a
dpm C(CH3)3 C(CH3)3 a
ppm C(CH3)3 CF »CF3 a
fod C(CH3)3 CF )CF CF3 a
acim CH3 H b
nona-F CF3 CH»CF3 b
acen CH3 - Cc

Ligand abbreviations for the structures shown in [link].

Attention has focused on Cu(II) B-diketonate [i.e., Cu(tfac)., Cu(hfac).] and
Cu(ID B-ketoiminate [i.e., Cu(acim), Cu(acen),]. An important characteristic of
Cu(II) compounds as CVD precursors is the use of heavily fluorinated ligand
such as Cu(tfac) and Cu(hfac)5 versus Cu(acac). The main effort of fluorine
substitution is a significant increase in the volatility of the complex.

Synthesis of Cu(II) precursors

Cu(hfac))"nH)O (n = 0, 1, 2)

Cu(hfac), is by far the most extensively studied of the Cu(II) CVD precursors.
Preparations in aqueous solutions yield the yellow-green dihydrate,
Cu(hfac) -2H,O. This is stable in very humid air or at lower temperatures but

slowly loses one molecule of water under typical laboratory conditions to form
the “grass-green” monohydrate, Cu(hfac)»-H»O. The monohydrate, which is
commercially available, can be sublimed unchanged and melts at 133 — 136 °C.
More vigorous drying over concentrated H»SO, produces the purple anhydrous
compound Cu(hfac), (mp = 95 — 98 °C). The purple material is hydroscopic,
converting readily into the monohydrate. Other B-diketonate Cu(II) complexes
are prepared by the similar method.

Schiff-base complexes

Schiff-base complexes include Cu(acim),, Cu(acen) and Cu(nona-F)>. The first
two of these can be prepared by mixing Cu(NH3),°* (aq) with the pure ligand
and by adding freshly prepared solid Cu(OH)> to a solution of the ligand in
acetone. The synthesis of Cu(nona-F),, on the other hand, involved two
important developments: the introduction of the silyl enol ether route to the
ligand and its conversion in-situ into the desired precursor. The new approach to
the ligand was required because, in contrast to non-fluorinated b-diketonates,
H(hfac) reacts with amines to produce salts.

Reaction mechanism

Starting from the experimental results, a list of possible steps for Cu CVD via
H, reduction of Cu(II) compounds would include the followings, where removal
of adsorbed ligand from the surface is believed to be the rate limiting step:

Cu(IDL>(g) > Cu(s) + 2 L-(ads)
H)(g) > 2 H(ads)
L-(ads) + H(ads) — HL(g)

where L represents any of the singly charged B-diketonate or B-ketoiminate
ligands described before. This mechanism gives a clear explanation of the
importance of hydrogen being present: in the absence of hydrogen, HL cannot
desorb cleanly into the gas phase and ligand will tend to decompose on the
surface, resulting in impurity incorporation into the growing film. The
mechanism is also supported by the observation that the deposition reaction is

enhanced by the addition of alcohol containing B-hydrogen to the reaction
mixture.

More recently, the focus has shifted to Cu(I) compounds including Cu(I)
cyclopentadienyls and Cu(I) B-diketonate. The Cu(I) B-diketonate in particular
show great promise as Cu CVD precursors and have superseded the Cu(II) B-
diketonate as the best family of precursors currently available.

From Cu(I) precursors

Precursor design

The Cu(I) compounds that have been investigated are described in [link]. These
species can be broadly divided into two classes, CuX and XCuL,, where X is a
uninegative ligand and L is a neutral Lewis base electron pair donor. The
XCuL, class can be further subdivided according to the nature of X and L.

Copper(I) Precursors

CuX a Vv
[Cu(OR)Iq a WN eke
[(RO)Cu(L)] , a an [ClCu(L)],
a
L
R} R? R' R’
7 R! R? nad
On 30 aa O._ LO
C Ree) Cu
I i fi
L

Copper(I) precursors used for CVD. Adapted from T. Kodas and
M. Hampden-Smith, The Chemistry of Metal CVD, VCH

Publishers Inc., New York, NY (1994).

Compounds of general formula CuX are likely to be oligomeric resulting in a
relatively low vapor pressure. The presence of a neutral donor ligand, L, is
likely to reduce the extent of oligomerization compared to CuX by occupying
vacant coordination sites. Metal alkoxide compounds are expected to undergo
thermal decomposition by cleavage of either M-O or O-C bonds.

Organo-copper(I) compounds, RCuL, where R is alkyl, are thermally unstable,
but cyclopentadienyl compounds are likely to be more robust due to the m-
bonding of the cyclopentadieny] ligand to the copper center. At the same time,
the cyclopentadieny] ligand is sterically demanding, occupies three coordination
sites at the metal center, and thereby reduces the desire for oligomerization. In
general, a cyclopentadieny] ligand is a poor choice to support CVD precursors,
especially with electropositive metals, because this ligand is unlikely to be
liable. Compounds in the family XCuL», where X is a halide and L is a
triorganophosphine, exhibit relatively high volatility but are thermally stable
with respect to formation of copper at low temperatures. These species are
therefore suitable as products of etching reactions of copper films.

A number of researchers have demonstrated the potential of a series of B-
diketonate Cu(I) compounds, (f-diketonate)CuL,, where L is Lewis base and n
= 1 or 2, that fulfill most of the criteria outlined for precursor design before.
These species were chosen as copper precursors for the following reasons:

e They contain the §-diketonate ligand which generally imparts volatility to
metal-organic complexes, particularly when fluorinated, as a result of a
reduction in hydrogen-bonding in the solid-state.

e They are capable of systematic substitution through both the B-diketonate
and Lewis base ligands to tailor volatility and reactivity.

e Lewis bases such as phosphines, olefins and alkynes are unlikely to
thermally decompose at temperatures where copper deposition occurs.

e These precursors can deposit copper via thermally induced
disproportionation reactions and no ligand decomposition is required since
the volatile Lewis base the Cu(II) disproportionation products are
transported out of the reactor intact at the disproportionation temperature.

Reaction mechanism

A general feature of the reactions of Cu(I) precursors is that they thermally
disproportionate, a mechanism likely to be responsible for the high purity of the
copper films observed since ligand decomposition does not occur. The
disproportionation mechanism is shown in [link] for (B-diketonate)CuL. The
unique capabilities of this class of compounds result from this reaction
mechanism by which they deposit copper. This mechanism is based on the
dissociative adsorption of the precursor to form Cu(hfac) and L,
disproportionation to form Cu(hfac), and Cu and desorption of Cu(hfac), and L.

San desorption

transport
to surface

ligand

Cu
un ee
Q dissociation iat. oo
C or —_— CS cy
| |

| |

substrate

Schematic diagram of the disproportionation mechanism. Adapted
from T. Kodas and M. Hampden-Smith, The Chemistry of Metal
CVD, VCH Publishers Inc., New York, NY (1994).

ee

Thus, the starting material acts as its own reducing agent and no external
reducing agent such as H> is required. Another advantage of the Cu(I) B-
diketonates over the Cu(II) B-diketonates is that in the former the ligand L can
be varied systematically, allowing the synthesis of a whole series of different but
closely related compounds.

Selectivity

Selectivity deposition has been studied in both hot- and cold-wall CVD reactors
as a function of the nature of the substrate, the temperature of the substrate and
the nature of the copper substituents. Selectivity has usually been evaluated by
using Si substrates on which SiO, has been grown and patterned with various
metals by either electron-beam deposition, CVD or sputtering. Research has
suggested that selectivity on metallic surfaces is attributable to the biomolecular
disproportionation reaction involved in precursor decomposition.

Bibliography

e J.R. Creighton, and J. E. Parmeter, Critical Review in Solid State and
Materials Science, 1993, 18, 175.

e L.H. Dubois and B. R. Zegarski, J. Electrochem. Soc., 1992, 139, 3295.

e J.J. Jarvis, R. Pearce, and M. F. Lappert, J. Chem. Soc., Dalton Trans.,
1977, 999.

e A. E. Kaloyeros, A. Feng, J. Garhart, K. C. Brooks, S. K. Ghosh, A. N.
Sazena, and F. Luehers, J. Electronic Mater., 1990, 19, 271.

e T. Kodas and M. Hampden-Smith, The Chemistry of Metal CVD, VCH
Publishers Inc., New York, NY (1994).

e C. F. Powell, J. H. Oxley, and J. M. Blocher Jr., Vapor Deposition, John
Wiley, New York (1966).

e S. Shingubara, Y. Nakasaki, and H. Kaneko, Appl. Phys. Lett., 1991, 58,
42.

Rutherford Backscattering of Thin Films

Introduction

One of the main research interests of the semiconductor industry is to improve the
performance of semiconducting devices and to construct new materials with reduced size or
thickness that have potential application in transistors and microelectronic devices.
However, the most significant challenge regarding thin film semiconductor materials is
measurement. Properties such as the thickness, composition at the surface, and
contamination, all are critical parameters of the thin films. To address these issues, we need
an analytical technique which can measure accurately through the depth of the of the
semiconductor surface without destruction of the material. Rutherford backscattering
spectroscopy is a unique analysis method for this purpose. It can give us information
regarding in-depth profiling in a non-destructive manner. However X-ray photo electron
spectroscopy (XPS), energy dispersive X-ray analysis (EDX) and Auger electron
spectroscopy are also able to study the depth-profile of semiconductor films. [link]
demonstrates the comparison between those techniques with RBS.

Method Hestuacave Incident Outgoing Detection Depth

particle Particle limit resolution

RBS No Ton Ion al 10 nm
X-ray

XPS Yes Electron ~0.1-1 ~1 pm
photon

EDX Yes Electron ale ~0.1 1.5 nm

photon
Auger Yes Electron Electron ~0.1-1 1.5 nm

Comparison between different thin film analysis techniques.

Basic concept of Rutherford backscattering spectroscopy

At a basic level, RBS demonstrates the electrostatic repulsion between high energy incident
ions and target nuclei. The specimen under study is bombarded with monoenergetic beam of
4He* particles and the backscattered particles are detected by the detector-analysis system
which measures the energies of the particles. During the collision, energy is transferred from

the incident particle to the target specimen atoms; the change in energy of the scattered
particle depends on the masses of incoming and target atoms. For an incident particle of
mass My, the energy is Eg while the mass of the target atom is M>. After the collision, the
residual energy E of the particle scattered at angle @ can be expressed as:

E= k 2E,

(sicoso + [M2 -M:?sin? @

Mit+Ma2

k=

where k is the kinematic scattering factor, which is actually the energy ratio of the particle
before and after the collision. Since k depends on the masses of the incident particle and
target atom and the scattering angle, the energy of the scattered particle is also determined
by these three parameters. A simplified layout of backscattering experiment is shown in
Figure 1.

Target

Incident ion

Scattering angle @

Detector

Scattered particle

Schematic representation of the experimental setup for
Rutherford backscattering analysis.

The probability of a scattering event can be described by the differential scattering cross
section of a target atom for scattering an incoming particle through the angle @ into
differential solid angle as follows,

M1.
doR | — [-os0+ Ft = Gagsinoy2|2

dp \2E0sin2@
1— eS sin@) 2

where dog is the effective differential cross section for the scattering of a particle. The
above equation may looks complicated but it conveys the message that the probability of
scattering event can be expressed as a function of scattering cross section which is
proportional to the zZ when a particle with charge ze approaches the target atom with
charge Ze.

Helium ions not scattered at the surface lose energy as they traverse the solid. They lose
energy due to interaction with electrons in the target. After collision the He particles lose
further energy on their way out to the detector. We need to know two quantities to measure
the energy loss, the distance At that the particles penetrate into the target and the energy loss
AE in this distance [link]. The rate of energy loss or stopping power is a critical component
in backscattering experiments as it determines the depth profile in a given experiment.

Depth

Target

Components of energy loss for a ion beam that
scatters from depth t. First, incident beam loses
energy through interaction with electrons AFjp.
Then energy lost occurs due to scattering E,.
Finally outgoing beam loses energy for interaction
with electrons AE,,,. Adapted from L. C. Feldman
and J. W. Mayer, Fundamentals of Surface and
Thin Film Analysis , North Holland-Elsevier, New
York (1986).

In thin film analysis, it is convenient to assume that total energy loss AE into depth t is only
proportional to t for a given target. This assumption allows a simple derivation of energy

loss in backscattering as more complete analysis requires many numerical techniques. In
constant dE/dx approximation, total energy loss becomes linearly related to depth t, [link].

Energy loss (AE)

Thickness “1000 A

Variation of energy loss with the depth of the target in
constant dE/dx approximation.

Experimental set-up

The apparatus for Rutherford backscattering analysis of thin solid surface typically consist
of three components:

1. A source of helium ions.
2. An accelerator to energize the helium ions.
3. A detector to measure the energy of scattered ions.

There are two types of accelerator/ion source available. In single stage accelerator, the He*
source is placed within an insulating gas-filled tank ({link]). It is difficult to install new ion
source when it is exhausted in this type of accelerator. Moreover, it is also difficult to
achieve particles with energy much more than 1 MeV since it is difficult to apply high
voltages in this type of system.

Source

Acceleration Tube |

i

Accelerated He* at 1 MeV

Schematic representation of a single stage accelerator.

Another variation is “tandem accelerator.” Here the ion source is at ground and produces
negative ion. The positive terminal is located is at the center of the acceleration tube
([link]). Initially the negative ion is accelerated from ground to terminal. At terminal two-
electron stripping process converts the He to He**. The positive ions are further accelerated
toward ground due to columbic repulsion from positive terminal. This arrangement can
achieve highly accelerated He** ions (~ 2.25 MeV) with moderate voltage of 750 kV.

Negative ion source

Acceleration Tube 4

Accelerated He* over 2 MeV

Schematic representation of a tandem accelerator.

Particles that are backscattered by surface atoms of the bombarded specimen are detected by
a surface barrier detector. The surface barrier detector is a thin layer of p-type silicon on the
n-type substrate resulting p-n junction. When the scattered ions exchange energy with the
electrons on the surface of the detector upon reaching the detector, electrons get promoted
from the valence band to the conduction band. Thus, each exchange of energy creates
electron-hole pairs. The energy of scattered ions is detected by simply counting the number
of electron-hole pairs. The energy resolution of the surface barrier detector in a standard
RBS experiment is 12 - 20 keV. The surface barrier detector is generally set between 90°
and 170° to the incident beam. Films are usually set normal to the incident beam. A simple
layout is shown in [link].

Incident Beam

Detector

Scattered Beam

Thin Film

~165°

Schematic representation general
setup where the surface barrier
detector is placed at angle of 165° to
the extrapolated incident beam.

Depth profile analysis

As stated earlier, it is a good approximation in thin film analysis that the total energy loss
AE is proportional to depth t. With this approximation, we can derive the relation between
energy width AE of the signal from a film of thickness At as follows,

AE = At(k dE/dx ;, + 1/cos@ dE/dx ou: )
where @ = lab scattering angle.

It is worth noting that k is the kinematic factor defined in equation above and the subscripts
“in” and “out” indicate the energies at which the rate of loss of energy or dE/dx is evaluated.
As an example, we consider the backscattering spectrum, at scattering angle 170°, for 2
MeV He" incidents on silicon layer deposited onto 2 mm thick niobium substrate [link].

400

100

Si layer on Nb substrate “4

0.6 0.8 1.0 £2 1.4 1.6 1.8 2.0
Energy (MeV)

The backscattering spectrum for 2.0 MeV He ions incident on a
silicon thin film deposited onto a niobium substrate. Adapted from
P. D. Stupik, M. M. Donovan, A. R. Barron, T. R. Jervis and M.
Nastasi, Thin Solid Films, 1992, 207, 138.

The energy loss rate of incoming He** or dE/dx along inward path in elemental Si is *24.6
eV/A at 2 MeV and is ©26 eV/A for the outgoing particle at 1.12 MeV (Since K of Si is
0.56 when the scattering angle is 170°, energy of the outgoing particle would be equal to 2 x
0.56 or 1.12 MeV) . Again the value of AE<; is ¥133.3 keV. Putting the values into above
equation we get

At © 133.3 keV/(0.56 * 24.6 eV/A + 1/cos 170° * 26 eV/A)

= 133.3 keV/(13.77 eV/A + 29/.985 eV/A)

= 133.3 keV/ 40.17 eV/A

= 3318 A.

Hence a Si layer of ca. 3300 A thickness has been deposited on the niobium substrate.
However we need to remember that the value of dE/dx is approximated in this calculation.

Quantitative Analysis

In addition to depth profile analysis, we can study the composition of an element
quantitatively by backscattering spectroscopy. The basic equation for quantitative analysis is

Y=o0.Q.Q. NAt

Where Y is the yield of scattered ions from a thin layer of thickness At, Q is the number of
incident ions and Q is the detector solid angle, and NAt is the number of specimen atoms
(atom/cm?). [link] shows the RBS spectrum for a sample of silicon deposited on a niobium
substrate and subjected to laser mixing. The Nb has reacted with the silicon to form a NbSi,
interphase layer. The Nb signal has broadened after the reaction as show in [link].

400

Spectrum after the formation of niobium silicide

Spectrum as Si deposited on Nb substrate

100 ;
NbSi_, layer on Nb |

AN

0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0
Energy (MeV)

Backscattering spectra of Si diffused into Nb and Si as deposited on
Nb substrate. Adapted from P. D. Stupik, M. M. Donovan, A. R.
Barron, T. R. Jervis and M. Nastasi, Thin Solid Films, 1992, 207,

138,

We can use ratio of the heights Hs;/Hyp of the backscattering spectrum after formation of
NbSi, to determine the composition of the silicide layer. The stoichiometric ratio of Nb and
Si can be approximated as,

Ngi/Nnp © LHs; * si ]/LHNb * Onp]

Hence the concentration of Si and Nb can be determined if we can know the appropriate
cross sections 0s; and Oj. However the yield in the backscattering spectra is better
represented as the product of signal height and the energy width AE. Thus stoichiometric
ratio can be better approximated as

Ngi/Nnp * [Hi * AEs; * osi]/LH np * AENp * Onp]

Limitations

It is of interest to understand the limitations of the backscattering technique in terms of the
comparison with other thin film analysis technique such as AES, XPS and SIMS ([link]).
AES has better mass resolution, lateral resolution and depth resolution than RBS. But AES
suffers from sputtering artifacts. Compared to RBS, SIMS has better sensitivity. RBS does
not provide any chemical bonding information which we can get from XPS. Again,
sputtering artifact problems are also associated in XPS. The strength of RBS lies in
quantitative analysis. However, conventional RBS systems cannot analyze ultrathin films
since the depth resolution is only about 10 nm using surface barrier detector.

Summary

Rutherford Backscattering analysis is a straightforward technique to determine the thickness
and composition of thin films (< 4000 A). Areas that have been lately explored are the use
of backscattering technique in composition determination of new superconductor oxides;
analysis of lattice mismatched epitaxial layers, and as a probe of thin film morphology and
surface clustering.

Bibliography

e L.C. Feldman and J. W. Mayer, Fundamentals of Surface and Thin Film Analysis,
North Holland-Elsevier, New York (1986).

e Ion Spectroscopies for Surface Analysis, Ed. A. W. Czanderna and D. M. Hercules,
Plenum Press (New York), 1991.

e P. D. Stupik, M. M. Donovan, A. R Barron, T. R. Jervis, and M. Nastasi, Thin Solid
Films, 1992, 207, 138

The Application of VSI (Vertical Scanning Interferometry) to the Study of
Crystal Surface Processes

Introduction

The processes which occur at the surfaces of crystals depend on many
external and internal factors such as crystal structure and composition,
conditions of a medium where the crystal surface exists and others. The
appearance of a crystal surface is the result of complexity of interactions
between the crystal surface and the environment. The mechanisms of
surface processes such as dissolution or growth are studied by the physical
chemistry of surfaces. There are a lot of computational techniques which
allows us to predict the changing of surface morphology of different
minerals which are influenced by different conditions such as temperature,
pressure, pH and chemical composition of solution reacting with the
surface. For example, Monte Carlo method is widely used to simulate the
dissolution or growth of crystals. However, the theoretical models of
surface processes need to be verified by natural observations. We can
extract a lot of useful information about the surface processes through
studying the changing of crystal surface structure under influence of
environmental conditions. The changes in surface structure can be studied
through the observation of crystal surface topography. The topography can
be directly observed macroscopically or by using microscopic techniques.
Microscopic observation allows us to study even very small changes and
estimate the rate of processes by observing changing the crystal surface
topography in time.

Much laboratory worked under the reconstruction of surface changes and
interpretation of dissolution and precipitation kinetics of crystals. Invention
of AFM made possible to monitor changes of surface structure during
dissolution or growth. However, to detect and quantify the results of
dissolution processes or growth it is necessary to determine surface area
changes over a significantly larger field of view than AFM can provide.
More recently, vertical scanning interferometry (VSI) has been developed
as new tool to distinguish and trace the reactive parts of crystal surfaces.
VSI and AFM are complementary techniques and practically well suited to
detect surface changes.

VSI technique provides a method for quantification of surface topography
at the angstrom to nanometer level. Time-dependent VSI measurements can
be used to study the surface-normal retreat across crystal and other solid
surfaces during dissolution process. Therefore, VSI can be used to directly
and nondirectly measure mineral dissolution rates with high precision.
Analogically, VSI can be used to study kinetics of crystal growth.

Physical principles of optical interferometry

Optical interferometry allows us to make extremely accurate measurements
and has been used as a laboratory technique for almost a hundred years.
Thomas Young observed interference of light and measured the wavelength
of light in an experiment, performed around 1801. This experiment gave an
evidence of Young's arguments for the wave model for light. The discovery
of interference gave a basis to development of interferomertry techniques
widely successfully used as in microscopic investigations, as in astronomic
investigations.

The physical principles of optical interferometry exploit the wave properties
of light. Light can be thought as electromagnetic wave propagating through
space. If we assume that we are dealing with a linearly polarized wave
propagating in a vacuum in z direction, electric field E can be represented
by a sinusoidal function of distance and time.

Equation:

E(x,y,z,t) = acos|2n(vt — z/A)|

Where a is the amplitude of the light wave, v is the frequency, and J is its
wavelength. The term within the square brackets is called the phase of the
wave. Let’s rewrite this equation in more compact form,

Equation:

E(a,y,z,t) = acos[wt — kz]

where w = 2rv is the circular frequency, and k = 2r/A is the propagation
constant. Let’s also transform this second equation into a complex

exponential form,
Equation:

E(x,y,z,t) = Re{aexp(id)exp(iwt)} = Re{ Aexp(iwt) }

where @ = 2nz/X and A = exp(—i¢@) is known as the complex amplitude.
If n is a refractive index of a medium where the light propagates, the light
wave traverses a distance d in such a medium. The equivalent optical path
in this case is

Equation:

p=n-d

When two light waves are superposed, the result intensity at any point
depends on whether reinforce or cancel each other ({link]). This is well
known phenomenon of interference. We will assume that two waves are
propagating in the same direction and are polarized with their field vectors
in the same plane. We will also assume that they have the same frequency.
The complex amplitude at any point in the interference pattern is then the
sum of the complex amplitudes of the two waves, so that we can write,
Equation:

A=A,+ Ap.

where A, = ayexp(—i¢,) and Az = a,exp(—id,) are the complex
amplitudes of two waves. The resultant intensity is, therefore,
Equation:

L= | A |? =y+Io+ 2(I,I2)/*cosAd

where J; and J2 are the intensities of two waves acting separately, and
Ad = ¢1 — ¢¢ is the phase difference between them. If the two waves are
derived from a common source, the phase difference corresponds to an
optical path difference,

Equation:

Ap = (A/2n) Ad

Two waves in phase

Interfering waves coming
from two point sources

Two waves out of phase

rt,

The scheme of interferometric wave interaction when two
waves interact with each other, the amplitude of resulting
wave will increase or decrease. The value of this amplitude
depends on phase difference between two original waves.

If Ad, the phase difference between the beams, varies linearly across the
field of view, the intensity varies cosinusoidally, giving rise to alternating
light and dark bands or fringes ([link]). The intensity in an interference
pattern has its maximum value

Equation:

Imax = Th + Ip + 2(IyI2)¥?

when Ad = 2mn, where m is an integer and its minimum value
Equation:

Tin = Ty + Ty — 2(E, IQ)?

when Ad = (2m + 1)z.

The principle of interferometry is widely used to develop many types of
interferometric set ups. One of the earliest set ups is Michelson
interferometry. The idea of this interferometry is quite simple: interference
fringes are produced by splitting a beam of monochromatic light so that one
beam strikes a fixed mirror and the other a movable mirror. An interference
pattern results when the reflected beams are brought back together. The
Michelson interferometric scheme is shown in [link].

Fixed mirror

Movable mirror

Incident beam

Detector

Schematic representation of a Michelson interferometry
set-up.

The difference of path lengths between two beams is 2x because beams
traverse the designated distances twice. The interference occurs when the
path difference is equal to integer numbers of wavelengths,

Equation:

ApS 2s => mA,m= Oye 1, = 2a

Modern interferometric systems are more complicated. Using special phase-
measurement techniques they capable to perform much more accurate
height measurements than can be obtained just by directly looking at the
interference fringes and measuring how they depart from being straight and
equally spaced. Typically interferometric system consist of lights source,
beamsplitter, objective system, system of registration of signals and
transformation into digital format and computer which process data.
Vertical scanning interferometry is contains all these parts. [link] shows a
configuration of VSI interferometric system.

Frame grabber

Computer

Magnification
selector

White Light Source

Beam Splitter

Microscope objec

pam

ail

PZT controller

Mireau objective) Reference mirror

Plate beamsplitte

\/ Sample Surface

Schematic representation of the Vertical scanning interferometry
(VSI) system.

Many of modern interferometric systems use Mirau objective in their
constructions. Mireau objective is based on a Michelson interferometer.
This objective consists of a lens, a reference mirror and a beamsplitter. The
idea of getting interfering beams is simple: two beams (red lines) travel
along the optical axis. Then they are reflected from the reference surface
and the sample surface respectively (blue lines). After this these beams are

recombined to interfere with each other. An illumination or light source
system is used to direct light onto a sample surface through a cube beam
splitter and the Mireau objective. The sample surface within the field of
view of the objective is uniformly illuminated by those beams with different
incidence angles. Any point on the sample surface can reflect those incident
beams in the form of divergent cone. Similarly, the point on the reference
symmetrical with that on the sample surface also reflects those illuminated
beams in the same form.

The Mireau objective directs the beams reflected of the reference and the
sample surface onto a CCD (charge-coupled device) sensor through a tube
lens. The CCD sensor is an analog shift register that enables the
transportation of analog signals (electric charges) through successive stages
(capacitors), controlled by a clock signal. The resulting interference fringe
pattern is detected by CCD sensor and the corresponding signal is digitized
by a frame grabber for further processing with a computer.

The distance between a minimum and a maximum of the interferogram
produced by two beams reflected from the reference and sample surface is
known. That is, exactly half the wavelength of the light source. Therefore,
with a simple interferogram the vertical resolution of the technique would
be also limited to A/2. If we will use a laser light as a light source with a
wavelength of 300 nm the resolution would be only 150 nm. This resolution
is not good enough for a detailed near-atomic scale investigation of crystal
surfaces. Fortunately, the vertical resolution of the technique can be
improved significantly by moving either the reference or the sample by a
fraction of the wavelength of the light. In this way, several interferograms
are produced. Then they are all overlayed, and their phase shifts compared
by the computer software [link]. This method is widely known as phase
shift interferometry (PSI).

Multiple interferograms
of the surface

FF

Se

Time 1} |

Sample surface

Resulting image
of the surface

Sketch illustrating phase-shift technology. The sample is
continuously moved along the vertical axes in order to scan surface
topography. All interferograms are automatically overlayed using
computer software.

Most optical testing interferometers now use phase-shifting techniques not
only because of high resolution but also because phase-shifting is a high
accuracy rapid way of getting the interferogram information into the
computer. Also usage of this technique makes the inherent noise in the data
taking process very low. As the result in a good environment angstrom or
sub-angstrom surface height measurements can be performed. As it was
said above, in phase-shifting interferometry the phase difference between
the interfering beams is changed at a constant rate as the detector is read
out. Once the phase is determined across the interference field, the

corresponding height distribution on the sample surface can be determined.
The phase distribution @(x, y) is recorded by using the CCD camera.

Let’s assign A(x, y), B(x, y), C(x, y) and D(x, y) to the resulting interference
light intensities which are corresponded to phase-shifting steps of 0, 7/2, 7
and 37/2. These intensities can be obtained by moving the reference mirror
through displacements of 1/8, A/4 and 3//8, respectively. The equations for
the resulting intensities would be:

Equation:

A(az,y) = I1(x,y) + Io(z,y)cosa(x,y)
Equation:

B(z,y) = Ii(2,y) — In(z,y)sina(z,y)
Equation:

C(z,y) = l(«,y) — In(x,y)cosa(x,y)
Equation:

D(a,y) = I1(x,y) + Io(x,y)sina(z,y)

where I; (x,y)and I2(x,y) are two overlapping beams from two symmetric
points on the test surface and the reference respectively. Solving equations
[link]—[link], the phase map @(x, y) of a sample surface will be given by the
relation:

Equation:

Once the phaseis determined across the interference field pixel by pixel on
a two-dimensional CCD array, the local height distribution/contour, h(x, y),
on the test surface is given by

Equation:

Normally the resulted fringe can be in the form of a linear fringe pattern by
adjusting the relative position between the reference mirror and sample
surfaces. Hence any distorted interference fringe would indicate a local
profile/contour of the test surface.

It is important to note that the Mireau objective is mounted on a capacitive
closed-loop controlled PZT (piezoelectric actuator) as to enable phase
shifting to be accurately implemented. The PZT is based on piezoelectric
effect referred to the electric potential generated by applying pressure to
piezoelectric material. This type of materials is used to convert electrical
energy to mechanical energy and vice-versa. The precise motion that results
when an electric potential is applied to a piezoelectric material has an
importance for nanopositioning. Actuators using the piezo effect have been
commercially available for 35 years and in that time have transformed the
world of precision positioning and motion control.

Vertical scanning interferometer also has another name; white-light
interferometry (WLI) because of using the white light as a source of light.
With this type of source a separate fringe system is produced for each
wavelength, and the resultant intensity at any point of examined surface is
obtained by summing these individual patterns. Due to the broad bandwidth
of the source the coherent length L of the source is short:

Equation:

2
L=
nAr

where A is the center wavelength, n is the refractive index of the medium,
AA is the spectral width of the source. In this way good contrast fringes can
be obtained only when the lengths of interfering beams pathways are closed
to each other. If we will vary the length of a pathway of a beam reflected
from sample, the height of a sample can be determined by looking at the
position for which a fringe contrast is a maximum. In this case interference

pattern exist only over a very shallow depth of the surface. When we vary a
pathway of sample-reflected beam we also move the sample in a vertical
direction in order to get the phase at which maximum intensity of fringes
will be achieved. This phase will be converted in height of a point at the
sample surface.

The combination of phase shift technology with white-light source provides
a very powerful tool to measure the topography of quite rough surfaces with
the amplitude in heights about and the precision up to 1-2 nm. Through a
developed software package for quantitatively evaluating the resulting
interferogram, the proposed system can retrieve the surface profile and
topography of the sample objects [link].

67.5 um

Example of muscovite surface topography, obtained by using VSI- 50x

objective.

A comparison of common methods to determine surface
topography: SEM, AFM and VSI

Except the interferometric methods described above, there are a several
other microscopic techniques for studying crystal surface topography. The
most common are scanning electron microscopy (SEM) and atomic force
microscopy (AFM). All these techniques are used to obtain information
about the surface structure. However they differ from each other by the
physical principles on which they based.

Scanning electron microscopy

SEM allows us to obtain images of surface topography with the resolution
much higher than the conventional light microscopes do. Also it is able to
provide information about other surface characteristics such as chemical
composition, electrical conductivity etc, see [link]. All types of data are
generated by the reflecting of accelerated electron beams from the sample
surface. When electrons strike the sample surface, they lose their energy by
repeated random scattering and adsorption within an outer layer into the
depth varying from 100 nm to 5 microns.

Incident beam

X-rays

Chemical composition of the surface Elastic backscattered electrons

Atomic number and toporgaphy

Cathodoluminescence Inelastic backscattered electrons
Electrical information Surface toporgaphy

NZ

Scheme of electron beam-sample interaction at SEM analysis

The thickness of this outer layer also knows as interactive layer depends on
energy of electrons in the beam, composition and density of a sample.
Result of the interaction between electron beam and the surface provides
several types of signals. The main type is secondary or inelastic scattered
electrons. They are produced as a result of interaction between the beam of
electrons and weakly bound electrons in the conduction band of the sample.
Secondary electrons are ejected from the k-orbitals of atoms within the
surface layer of thickness about a few nanometers. This is because
secondary electrons are low energy electrons (<50 eV), so only those
formed within the first few nanometers of the sample surface have enough
energy to escape and be detected. Secondary backscattered electrons
provide the most common signal to investigate surface topography with
lateral resolution up to 0.4 - 0.7 nm.

High energy beam electrons are elastic scattered back from the surface. This
type of signal gives information about chemical composition of the surface
because the energy of backscattered electrons depends on the weight of
atoms within the interaction layer. Also this type of electrons can form
secondary electrons and escape from the surface or travel father into the
sample than the secondary. The SEM image formed is the result of the
intensity of the secondary electron emission from the sample at each x,y
data point during the scanning of the surface.

Atomic force microscopy

AFM is a very popular tool to study surface dissolution. AFM set up
consists of scanning a sharp tip on the end of a flexible cantilever which
moves across a sample surface. The tips typically have an end radius of 2 to
20 nm, depending on tip type. When the tip touch the surface the forces of
these interactions leads to deflection of a cantilever. The interaction
between tip and sample surface involve mechanical contact forces, van der
Waals forces, capillary forces, chemical bonding, electrostatic forces,
magnetic forces etc. The deflection of a cantilever is usually measured by
reflecting a laser beam off the back of the cantilever into a split photodiode
detector. A schematic drawing of AFM can be seen in [link]. The two most
commonly used modes of operation are contact mode AFM and tapping
mode AFM, which are conducted in air or liquid environments.

Photodetector
Laser beam

Cantilever

Schematic drawing of an AFM apparatus.

Working under the contact mode AFM scans the sample while monitoring
the change in cantilever deflection with the split photodiode detector. Loop
maintains a constant cantilever reflection by vertically moving the scanner

to get a constant signal. The distance which the scanner goes by moving
vertically at each x,y data point is stored by the computer to form the
topographic image of the sample surface. Working under the tapping mode
AFM oscillates the cantilever at its resonance frequency (typically~300
kHz) and lightly “taps” the tip on the surface during scanning. The
electrostatic forces increase when tip gets close to the sample surface,
therefore the amplitude of the oscillation decreases. The laser deflection
method is used to detect the amplitude of cantilever oscillation. Similar to
the contact mode, feedback loop maintains a constant oscillation amplitude
by moving the scanner vertically at every x,y data point. Recording this
movement forms the topographical image. The advantage of tapping mode
over contact mode is that it eliminates the lateral, shear forces present in
contact mode. This enables tapping mode to image soft, fragile, and
adhesive surfaces without damaging them while work under contact mode
allows the damage to occur.

Comparison of techniques

All techniques described above are widely used in studying of surface nano-
and micromorphology. However, each method has its own limitations and
the proper choice of analytical technique depends on features of analyzed
surface and primary goals of research.

All these techniques are capable to obtain an image of a sample surface
with quite good resolution. The lateral resolution of VSI is much less, then
for other techniques: 150 nm for VSI and 0.5 nm for AFM and SEM.
Vertical resolution of AFM (0.5 A) is better then for VSI (1 - 2 nm),
however VSI is capable to measure a high vertical range of heights (1 mm)
which makes possible to study even very rough surfaces. On the contrary,
AFM allows us to measure only quite smooth surfaces because of its
relatively small vertical scan range (7 pm). SEM has less resolution, than
AFM because it requires coating of a conductive material with the thickness
within several nm.

The significant advantage of VSI is that it can provide a large field of view
(845 x 630 pm for 10x objective) of tested surfaces. Recent studies of

surface roughness characteristics showed that the surface roughness
parameters increase with the increasing field of view until a critical size of
250,000 pm is reached. This value is larger then the maximum field of view
produced by AFM (100 x 100 pm) but can be easily obtained by VSI. SEM
is also capable to produce images with large field of view. However, SEM
is able to provide only 2D images from one scan while AFM and VSI let us
to obtain 3D images. It makes quantitative analysis of surface topography
more complicated, for example, topography of membranes is studied by
cross section and top view images.

VSI AFM SEM
Palle 0.5-1.2 um sa 0.5-1 nm
resolution nm
Mewes 2nm 05A Only 2D images
resolution
’ 845 x 630 pm 100 x
meu (10x 100 1-2 mm
view at
objective) pm
Vertical
10
range of 1mm -
pm
scan
Preparation ; : Required coating of
of a sample a conducted material
Required Air Alr, . Vacuum
environment liquid

A comparison of VSI sample and resolution with AFM and SEM.

The experimental studying of surface processes using
microscopic techniques

The limitations of each technique described above are critically important
to choose appropriate technique for studying surface processes. Let’s
explore application of these techniques to study dissolution of crystals.

When crystalline matter dissolves the changes of the crystal surface
topography can be observed by using microscopic techniques. If we will
apply an unreactive mask (silicon for example) on crystal surface and place
a crystalline sample into the experiment reactor then we get two types of
surfaces: dissolving and remaining the same or unreacted. After some
period of time the crystal surface starts to dissolve and change its z-level. In
order to study these changes ex situ we can pull out a sample from the
reaction cell then remove a mask and measure the average height difference
Ahbetween the unreacted and dissolved areas. The average heights of
dissolved and unreacted areas are obtained through digital processing of
data obtained by microscopes. The velocity of normal surface retreat Usnr
during the time interval At is defined as

Ah
USNR = At

Dividing this velocity by the molar volume V(cm*/mol) gives a global
dissolution rate in the familiar units of moles per unit area per unit time:
Equation:

USNR
V

Pe

This method allows us to obtain experimental values of dissolution rates
just by precise measuring of average surface heights. Moreover, using this
method we can measure local dissolution rates at etch pits by monitoring
changes in the volume and density of etch pits across the surface over time.
VSI technique is capable to perform these measurements because of large
vertical range of scanning. In order to get precise values of rates which are

not depend on observing place of crystal surface we need to measure
enough large areas. VSI technique provides data from areas which are large
enough to study surfaces with heterogeneous dissolution dynamics and
obtain average dissolution rates. Therefore, VSI makes possible to measure
rates of normal surface retreat during the dissolution and observe formation,
growth and distribution of etch pits on the surface.

However, if the mechanism of dissolution is controlled by dynamics of
atomic steps and kink sites within a smooth atomic surface area, the
observation of the dissolution process need to use a more precise technique.
AFM is capable to provide information about changes in step morphology
in situ when the dissolution occurs. For example, immediate response of the
dissolved surface to the changing of environmental conditions
(concentrations of ions in the solution, pH etc.) can be studied by using
AFM.

SEM is also used to examine micro and nanotexture of solid surfaces and
study dissolution processes. This method allows us to observe large areas of
crystal surface with high resolution which makes possible to measure a high
variety of surfaces. The significant disadvantage of this method is the
requirement to cover examine sample by conductive substance which limits
the resolution of SEM. The other disadvantage of SEM is that the analysis
is conducted in vacuum. Recent technique, environmental SEM or ESEM
overcomes these requirements and makes possible even examine liquids
and biological materials. The third disadvantage of this technique is that it
produces only 2D images. This creates some difficulties to measure Ah
within the dissolving area. One of advantages of this technique is that it is
able to measure not only surface topography but also chemical composition
and other surface characteristics of the surface. This fact is used to monitor
changing in chemical composition during the dissolution.

Bibliography

e A.C. Lasaga, Kinetic Theory in the Earth Sciences. Princeton Univ.
Press, Princeton, NJ (1998).

e A. Luttge, E. V. Bolton, and A. C. Lasaga A.C., Am. J. Sci., 1999, 299,
652.

D. Kaczmarek, Vacuum, 2001, 62, 303.

P. Hariharan. Optical interferometry, Second edition, Academic press
(2003) ISBN 0-12-311630-9.

A. Luttge and P. G. Conrad, Appl. Environ. Microbiol., 2004, 70, 1627.
A. C. Lasaga and A. Luttge, American Mineralogist, 2004, 89, 527.
K. J. Davis and A. Luttge, Am. J. Sci., 2005, 305, 727.

S. H. Wang and Tay, Meas. Sci. Technol., 2006, 17, 617.

A. Luttge and R. S. Arvidson, in Kinetics of water-rock interaction,
Ed. S. Brantley, J. Kubicki, and A. White, Springer (2007).

L. Zhang and A. Luttge, American Mineralogist, 2007, 92, 1316.

C. Fischer A. and Luttge, Am. J. Sci., 2007, 307, 955.

Y. Wyart, G. Georges, C. Deumie, C. Amra, and P. Moulina, J.
Membrane Sci., 2008, 315, 82.

T. C. Vaimakis, E. D. Economou, and C. C. Trapalis, J. Therm. Anal.
Cal., 2008, 92, 783.

Atomic Force Microscopy

Introduction

Atomic force microscopy (AFM) is a high-resolution form of scanning
probe microscopy, also known as scanning force microscopy (SFM). The
instrument uses a cantilever with a sharp tip at the end to scan over the
sample surface ({link]). As the probe scans over the sample surface,
attractive or repulsive forces between the tip and sample, usually in the
form of van der Waal forces but also can be a number of others such as
electrostatic and hydrophobic/hydrophilic, cause a deflection of the
cantilever. The deflection is measured by a laser ({link]) which is reflected
off the cantilever into photodiodes. As one of the photodiodes collects more
light, it creates an output signal that is processed and provides information
about the vertical bending of the cantilever. This data is then sent to a
scanner that controls the height of the probe as it moves across the surface.
The variance in height applied by the scanner can then be used to produce a
three-dimensional topographical representation of the sample.

Cantilever
with Tip

=>
Sample

Simple schematic of

atomic force microscope
(AFM) apparatus.
Adapted from H. G.
Hansma, Department of
Physics, University of
California, Santa Barbara.

Modes of operation

Contact mode

The contact mode method utilizes a constant force for tip-sample
interactions by maintaining a constant tip deflection ([link].). The tip
communicates the nature of the interactions that the probe is having at the
surface via feedback loops and the scanner moves the entire probe in order
to maintain the original deflection of the cantilever. The constant force is
calculated and maintained by using Hooke's Law, [link]. This equation
relates the force (F), spring constant (k), and cantilever deflection (x). Force
constants typically range from 0.01 to 1.0 N/m. Contact mode usually has
the fastest scanning times but can deform the sample surface. It is also only
the only mode that can attain "atomic resolution."

Equation:

F = -kx

Schematic diagram of

probe and surface
interaction in contact
mode.

Tapping mode

In the tapping mode the cantilever is externally oscillated at its fundamental
resonance frequency ({link]). A piezoelectric on top of the cantilever is used
to adjust the amplitude of oscillation as the probe scans across the surface.
The deviations in the oscillation frequency or amplitude due to interactions
between the probe and surface are measured, and provide information about
the surface or types of material present in the sample. This method is
gentler than contact AFM since the tip is not dragged across the surface, but
it does require longer scanning times. It also tends to provide higher lateral
resolution than contact AFM.

fia
ee

Diagram of probe and
surface interaction in
tapping mode.

Noncontact mode

For noncontact mode the cantilever is oscillated just above its resonance
frequency and this frequency is decreased as the tip approaches the surface
and experiences the forces associated with the material ((link]). The average
tip-to-sample distance is measured as the oscillation frequency or amplitude
is kept constant, which then can be used to image the surface. This method
exerts very little force on the sample, which extends the lifetime of the tip.
However, it usually does not provide very good resolution unless placed
under a strong vacuum.

Diagram of probe and
surface interaction in
noncontact mode.

Experimental limitations

A common problem seen in AFM images is the presence of artifacts which
are distortions of the actual topography, usually either due to issues with the
probe, scanner, or image processing. The AFM scans slowly which makes it
more susceptible to external temperature fluctuations leading to thermal
drift. This leads to artifacts and inaccurate distances between topographical
features.

It is also important to consider that the tip is not perfectly sharp and
therefore may not provide the best aspect ratio, which leads to a
convolution of the true topography. This leads to features appearing too
large or too small since the width of the probe cannot precisely move
around the particles and holes on the surface. It is for this reason that tips

with smaller radii of curvature provide better resolution in imaging. The tip
can also produce false images and poorly contrasted images if it is blunt or
broken.

The movement of particles on the surface due to the movement of the
cantilever can cause noise, which forms streaks or bands in the image.
Artifacts can also be made by the tip being of inadequate proportions
compared to the surface being scanned. It is for this reason that it is
important to use the ideal probe for the particular application.

Sample size and preparation

The sample size varies with the instrument but a typical size is 8 mm by 8
mm with a typical height of 1 mm. Solid samples present a problem for
AFM since the tip can shift the material as it scans the surface. Solutions or
dispersions are best for applying as uniform of a layer of material as
possible in order to get the most accurate value of particles’ heights. This is
usually done by spin-coating the solution onto freshly cleaved mica which
allows the particles to stick to the surface once it has dried.

Applications of AFM

AFM is particularly versatile in its applications since it can be used in
ambient temperatures and many different environments. It can be used in
many different areas to analyze different kinds of samples such as
semiconductors, polymers, nanoparticles, biotechnology, and cells amongst
others. The most common application of AFM is for morphological studies
in order to attain an understanding of the topography of the sample. Since it
is common for the material to be in solution, AFM can also give the user an
idea of the ability of the material to be dispersed as well as the homogeneity
of the particles within that dispersion. It also can provide a lot of
information about the particles being studied such as particle size, surface
area, electrical properties, and chemical composition. Certain tips are
capable of determining the principal mechanical, magnetic, and electrical
properties of the material. For example, in magnetic force microscopy
(MFM) the probe has a magnetic coating that senses magnetic, electrostatic,
and atomic interactions with the surface. This type of scanning can be

performed in static or dynamic mode and depicts the magnetic structure of
the surface.

AFM of carbon nanotubes

Atomic force microscopy is usually used to study the topographical
morphology of these materials. By measuring the thickness of the material
it is possible to determine if bundling occurred and to what degree. Other
dimensions of the sample can also be measured such as the length and
width of the tubes or bundles. It is also possible to detect impurities,
functional groups ({link]), or remaining catalyst by studying the images.
Various methods of producing nanotubes have been found and each
demonstrates a slightly different profile of homogeneity and purity. These
impurities can be carbon coated metal, amorphous carbon, or other
allotropes of carbon such as fullerenes and graphite. These facts can be
utilized to compare the purity and homogeneity of the samples made from
different processes, as well as monitor these characteristics as different
steps or reactions are performed on the material. The distance between the
tip and the surface has proven itself to be an important parameter in
noncontact mode AFM and has shown that if the tip is moved past the
threshold distance, approximately 30 pm, it can move or damage the
nanotubes. If this occurs, a useful characterization cannot be performed due
to these distortions of the image.

0
0 0.5 1.0 1.5 um

AFM image of a
polyethyleneimine-
functionalized single walled
carbon nanotube (PEI-SWNT)
showing the presence of PEI
“globules” on the SWNT.
Adapted from E. P. Dillon, C.
A. Crouse, and A. R. Barron,
ACS Nano, 2008, 2, 156.

AFM of fullerenes

Atomic force microscopy is best applied to aggregates of fullerenes rather
than individual ones. While the AFM can accurately perform height
analysis of individual fullerene molecules, it has poor lateral resolution and
it is difficult to accurately depict the width of an individual molecule.
Another common issue that arises with contact AFM and fullerene
deposited films is that the tip shifts clusters of fullerenes which can lead to
discontinuities in sample images.

Bibliography

e R. Anderson and A. R. Barron, J. Am. Chem. Soc., 2005, 127, 10458.

e M. Bellucci, G. Gaggiotti, M. Marchetti, F. Micciulla, R. Mucciato,
and M. Regi, J. Physics: Conference Series, 2007, 61, 99.

e J. I. Bobrinetskii, V. N. Kukin, V. K. Nevolin, and M. M. Simunin.
Semiconductor, 2008, 42, 1496.

e S.H. Cohen and M. L. Lightbody. Atomic Force Microscopy/Scanning
Tunneling Microscopy 2. Plenum, New York (1997).

e E. P. Dillon, C. A. Crouse, and A. R. Barron, ACS Nano, 2008, 2, 156.

e C. Gu, C. Ray, S. Guo, and B. B. Akhremitchev, J. Phys. Chem., 2007,
111, 12898.

e G. Kaupp, Atomic Force Microscopy, Scanning Nearfield Optical
Microscopy and Nanoscratching: Application to Rough and Natural
Surfaces. Springer-Verlag, Berlin (2006).

e S. Morita, R. Wiesendanger, E. Meyer, and F. J. Giessibl. Noncontact
Atomic Force Microscopy. Springer, Berlin (2002).

Introduction to Nanoparticle Synthesis

The fabrication of nanomaterials with strict control over size, shape, and
crystalline structure has inspired the application of nanochemistry to
numerous fields including catalysis, medicine, and electronics. The use of
nanomaterials in such applications also requires the development of
methods for nanoparticle assembly or dispersion in various media. A
majority of studies have been aimed at dispersion in aqueous media aimed
at their use in medical applications and studies of environmental effects,
however, the principles of nanoparticle fabrication and functionalization of
nanoparticles transcends their eventual application. Herein, we review the
most general routes to nanoparticles of the key types that may have
particular application within the oil and gas industry for sensor, composite,
or device applications.

Synthesis methods for nanoparticles are typically grouped into two
categories: “top-down” and “bottom-up”. The first involves division of a
massive solid into smaller portions. This approach may involve milling or
attrition, chemical methods, and volatilization of a solid followed by
condensation of the volatilized components. The second, “bottom-up”,
method of nanoparticle fabrication involves condensation of atoms or
molecular entities in a gas phase or in solution. The latter approach is far
more popular in the synthesis of nanoparticles.

Dispersions of nanoparticles are intrinsically thermodynamically
metastable, primarily due to their very high surface area, which represents a
positive contribution to the free enthalpy of the system. If the activation
energies are not sufficiently high, evolution of the nanoparticle dispersion
occurs causing an increase in nanoparticle size as typified by an Ostwald
ripening process. Thus, highly dispersed nanoparticles are only kinetically
stabilized and cannot be prepared under conditions that exceed some
threshold, meaning that so-called “soft-chemical” or “chemie duce”
methods are preferred. In addition, the use of surface stabilization is
employed in many nanomaterials to hinder sintering, recrystallization and
aggregation.

Bibliography

e J. Gopalakrishnan, Chem. Mater., 1995, 7, 1265.

Synthesis of Semiconductor Nanoparticles

The most studied non-oxide semiconductors are cadmium chalcogenides
(CdE, with E = sulfide, selenide and telluride). CdE nanocrystals were
probably the first material used to demonstrate quantum size effects
corresponding to a change in the electronic structure with size, i.e., the
increase of the band gap energy with the decrease in size of particles
({link]). These semiconductors nanocrystals are commonly synthesized by
thermal decomposition of an organometallic precursor dissolved in an
anhydrous solvent containing the source of chalcogenide and a stabilizing
material (polymer or capping ligand). Stabilizing molecules bound to the
surface of particles control their growth and prevent particle aggregation.

Picture of cadmium selenide
(CdSe) quantum dots, dissolved
in toluene, fluorescing brightly,

as they are exposed to an
ultraviolet lamp, in three
noticeable different colors (blue
~481 nm, green ~520 nm, and
orange ~612 nm) due to the
quantum dots' bandgap (and
thus the wavelength of emitted
light) depends strongly on the
particle size; the smaller the dot,

the shorter the emitted
wavelength of light. The "blue"
quantum dots have the smallest
particle size, the "green" dots
are slightly larger, and the
"orange" dots are the largest.

Although cadmium chalcogenides are the most studies semiconducting
nanoparticles, the methodology for the formation of semiconducting
nanoparticles was first demonstrated independently for InP and GaAs, e.g.,
[link]. This method has been adapted for a range of semiconductor
nanoparticles.

Equation:

InCl, + P(SiMe;); > InP + 3 Me,SiCl

In the case of CdE, dimethylcadmium Cd(CHs3), is used as a cadmium
source and bis(trimethylsilyl)sulfide, (Me3Si)2S, trioctylphosphine selenide
or telluride (TOPSe, TOPTe) serve as sources of selenide in
trioctylphosphine oxide (TOPO) used as solvent and capping molecule. The
mixture is heated at 230-260 °C over a few hours while modulating the
temperature in response to changes in the size distribution as estimated
from the absorption spectra of aliquots removed at regular intervals. These
particles, capped with TOP/TOPO molecules, are non-aggregated ([link])
and easily dispersible in organic solvents forming optically clear
dispersions. When similar syntheses are performed in the presence of
surfactant, strongly anisotropic nanoparticles are obtained, e.g., rod-shaped
CdSe nanoparticles can be obtained.

TEM image of CdSe
nanoparticles.

Because Cd(CH3)> is extremely toxic, pyrophoric and explosive at elevated
temperature, other Cd sources have been used. CdO appears to be an
interesting precursor. CdO powder dissolves in TOPO and HPA or TDPA
(tetradecylphosphonic acid) at about 300 °C giving a colorless
homogeneous solution. By introducing selenium or tellurium dissolved in
TOP, nanocrystals grow to the desired size.

Nanorods of CdSe or CdTe can also be produced by using a greater initial
concentration of cadmium as compared to reactions for nanoparticles. This
approach has been successfully applied for synthesis of numerous other
metal chalcogenides including ZnS, ZnSe, and Zn,_,Cd,S. Similar
procedures enable the formation of MnS, PdS, NiS, Cu»S nanoparticles,
nano rods, and nano disks.

Bibliography

e C.R. Berry, Phys. Rev., 1967, 161, 848.

M. D. Healy, P. E. Laibinis, P. D. Stupik, and A. R. Barron, J. Chem.
Soc., Chem. Commun., 1989, 359.

L. Manna, E. C. Scher, and A. P. Alivisatos, J. Am. Chem. Soc., 2000,
122, 12700.

C. B. Murray, D. J. Norris, and M. G. Bawendi, J. Am. Chem. Soc.,
1993, 115, 8706.

Z. A. Peng and X. Peng, J. Am. Chem. Soc., 2002, 12, 3343.

R. L. Wells, C. G. Pitt, A. T. McPhail, A. P. Purdy, S. R. B. Shafieezad,
and Hallock Chem. Mater., 1989, 1, 4.

X. Zong, Y. Feng, W. Knoll, and H. Man, J. Am. Chem. Soc., 2003,
125, 13559.

Optical Properties of Group 12-16 (II-VI) Semiconductor Nanoparticles

What are Group 12-16 semiconductors?

Semiconductor materials are generally classified on the basis of the periodic
table group that their constituent elements belong to. Thus, Group 12-16
semiconductors, formerly called II-VI semiconductors, are materials whose
cations are from the Group 12 and anions are from Group 16 in the periodic
table ([link]). Some examples of Group 12-16 semiconductor materials are
cadmium selenide (CdSe), zinc sulfide (ZnS), cadmium teluride (CdTe),
zinc oxide (ZnO), and mercuric selenide (HgSe) among others.

Note:The new IUPAC (International Union of Pure and Applied
Chemistry) convention is being followed in this document, to avoid any
confusion with regard to conventions used earlier. In the old IUPAC
convention, Group 12 was known as Group IIB with the roman numeral
‘II’ referring to the number of electrons in the outer electronic shells and B
referring to being on the right part of the table. However, in the CAS
(Chemical Abstracts Service), the alphabet B refers to transition elements
as compared to main group elements, though the roman numeral has the
same meaning. Similarly, Group 16 was earlier known as Group VI
because all the elements in this group have 6 valence shell electrons.

The red box
indicates the Group
12 and Group 16
elements in the
periodic table.

What are Group 12-16 (II-VI) semiconductor nanoparticles?

From the Greek word nanos - meaning "dwarf" this prefix is used in the
metric system to mean 10°° or one billionth (1/1,000,000,000). Thus a
nanometer is 10°? or one billionth of a meter, and a nanojoule is 10°? or one
billionth of a Joule, etc. A nanoparticle is ordinarily defined as any particle
with at least one of its dimensions in the 1 - 100 nm range.

Nanoscale materials often show behavior which is intermediate between
that of a bulk solid and that of an individual molecule or atom. An inorganic
nanocrystal can be imagined to be comprised of a few atoms or molecules.
It thus will behave differently from a single atom; however, it is still smaller
than a macroscopic solid, and hence will show different properties. For
example, if one would compare the chemical reactivity of a bulk solid anda
nanoparticle, the latter would have a higher reactivity due to a significant
fraction of the total number of atoms being on the surface of the particle.
Properties such as boiling point, melting point, optical properties, chemical
stability, electronic properties, etc. are all different in a nanoparticle as
compared to its bulk counterpart. In the case of Group 12-16
semiconductors, this reduction in size from bulk to the nanoscale results in
many size dependent properties such as varying band gap energy, optical
and electronic properties.

Optical properties of semiconductor quantum nanoparticles

In the case of semiconductor nanocrystals, the effect of the size on the
optical properties of the particles is very interesting. Consider a Group 12-
16 semiconductor, cadmium selenide (CdSe). A 2 nm sized CdSe crystal

has a blue color fluorescence whereas a larger nanocrystal of CdSe of about
6 nm has a dark red fluorescence ([link]). In order to understand the size
dependent optical properties of semiconductor nanoparticles, it is important
to know the physics behind what is happening at the nano level.

Fluorescing CdSe
quantum dots synthesized
in a heat transfer liquid of

different sizes (M. S.
Wong, Rice University).

Energy levels in a semiconductor

The electronic structure of any material is given by a solution of
Schrédinger equations with boundary conditions, depending on the physical
situation. The electronic structure of a semiconductor ({link]) can be
described by the following terms:

Conduction band

Band Gap =| Eg (bulk)

Valence band

Simplified
representation
of the energy
levels ina
bulk
semiconductor

Energy level

By the solution of Schrédinger’s equations, the electrons in a
semiconductor can have only certain allowable energies, which are
associated with energy levels. No electrons can exist in between these
levels, or in other words can have energies in between the allowed energies.
In addition, from Pauli’s Exclusion Principle, only 2 electrons with opposite
spin can exist at any one energy level. Thus, the electrons start filling from
the lowest energy levels. Greater the number of atoms in a crystal, the
difference in allowable energies become very small, thus the distance
between energy levels decreases. However, this distance can never be zero.
For a bulk semiconductor, due to the large number of atoms, the distance

between energy levels is very small and for all practical purpose the energy
levels can be described as continuous ([link]).

Band gap

From the solution of Schrédinger’s equations, there are a set of energies
which is not allowable, and thus no energy levels can exist in this region.
This region is called the band gap and is a quantum mechanical
phenomenon ({link]). In a bulk semiconductor the bandgap is fixed;
whereas in a quantum dot nanoparticle the bandgap varies with the size of
the nanoparticle.

Valence band

In bulk semiconductors, since the energy levels can be considered as
continuous, they are also termed as energy bands. Valence band contains
electrons from the lowest energy level to the energy level at the lower edge
of the bandgap ([link]). Since filling of energy is from the lowest energy
level, this band is usually almost full.

Conduction band

The conduction band consists of energy levels from the upper edge of the
bandgap and higher ([link]). To reach the conduction band, the electrons in
the valence band should have enough energy to cross the band gap. Once
the electrons are excited, they subsequently relax back to the valence band
(either radiatively or non-radiatively) followed by a subsequent emission of
radiation. This property is responsible for most of the applications of
quantum dots.

Exciton and exciton Bohr radius

When an electron is excited from the valence band to the conduction band,
corresponding to the electron in the conduction band a hole (absence of
electron) is formed in the valence band. This electron pair is called an
exciton. Excitons have a natural separation distance between the electron
and hole, which is characteristic of the material. This average distance is
called exciton Bohr radius. In a bulk semiconductor, the size of the crystal
is much larger than the exciton Bohr radius and hence the exciton is free to
move throughout the crystal.

Energy levels in a quantum dot semiconductor

Before understanding the electronic structure of a quantum dot
semiconductor, it is important to understand what a quantum dot
nanoparticle is. We earlier studied that a nanoparticle is any particle with
one of its dimensions in the 1 - 100 nm. A quantum dot is a nanoparticle
with its diameter on the order of the materials exciton Bohr radius.
Quantum dots are typically 2 - 10 nm wide and approximately consist of 10
to 50 atoms. With this understanding of a quantum dot semiconductor, the
electronic structure of a quantum dot semiconductor can be described by the
following terms.

Conduction band

2S(e)

1D(e)
1P(e)
1S(e)
E,(QD)
1S(h
1P(h) ©)

Valence band

Energy levels in

quantum dot.
Allowed optical
transitions are
shown. Adapted
from T. Pradeep,
Nano: The
Essentials.
Understanding
Nanoscience and
Nanotechnology,
Tata McGraw-Hill,
New Delhi (2007).

Quantum confinement

When the size of the semiconductor crystal becomes comparable or smaller
than the exciton Bohr radius, the quantum dots are in a state of quantum
confinement. As a result of quantum confinement, the energy levels in a
quantum dot are discrete ([link]) as opposed to being continuous in a bulk
crystal ([link]).

Discrete energy levels

In materials that have small number of atoms and are considered as
quantum confined, the energy levels are separated by an appreciable
amount of energy such that they are not continuous, but are discrete (see
[link]). The energy associated with an electron (equivalent to conduction
band energy level) is given by is given by [link], where h is the Planck’s
constant, mz is the effective mass of electron and n is the quantum number
for the conduction band states, and n can take the values 1, 2, 3 and so on.
Similarly, the energy associated with the hole (equivalent to valence band
energy level) is given by [link], where n' is the quantum number for the

valence states, and n’ can take the values 1, 2, 3, and so on. The energy
increases as one goes higher in the quantum number. Since the electron
mass is much smaller than that of the hole, the electron levels are separated
more widely than the hole levels.

Equation:
Fe = hen?
817m, a
Equation:
fs . Par
827m,
Tunable band gap

As seen from [link] and [link], the energy levels are affected by the
diameter of the semiconductor particles. If the diameter is very small, since
the energy is dependent on inverse of diameter squared, the energy levels of
the upper edge of the band gap (lowest conduction band level) and lower
edge of the band gap (highest valence band level) change significantly with
the diameter of the particle and the effective mass of the electron and the
hole, resulting in a size dependent tunable band gap. This also results in the
discretization of the energy levels.

Qualitatively, this can be understood in the following way. In a bulk
semiconductor, the addition or removal of an atom is insignificant
compared to the size of the bulk semiconductor, which consists of a large
number of atoms. The large size of bulk semiconductors makes the changes
in band gap so negligible on the addition of an atom, that it is considered as
a fixed band gap. In a quantum dot, addition of an atom does make a
difference, resulting in the tunability of band gap.

UV-visible absorbance

Due to the presence of discrete energy levels in a QD, there is a widening of
the energy gap between the highest occupied electronic states and the
lowest unoccupied states as compared to the bulk material. As a
consequence, the optical properties of the semiconductor nanoparticles also
become size dependent.

The minimum energy required to create an exciton is the defined by the
band gap of the material, i.e., the energy required to excite an electron from
the highest level of valence energy states to the lowest level of the
conduction energy states. For a quantum dot, the bandgap varies with the
size of the particle. From [link] and [link], it can be inferred that the band
gap becomes higher as the particle becomes smaller. This means that for a
smaller particle, the energy required for an electron to get excited is higher.
The relation between energy and wavelength is given by [link], where h is
the Planck’s constant, c is the speed of light, A is the wavelength of light.
Therefore, from [link] to cross a bandgap of greater energy, shorter
wavelengths of light are absorbed, i.e., a blue shift is seen.

Equation:

B= he
ny

For Group 12-16 semiconductors, the bandgap energy falls in the UV-
visible range. That is ultraviolet light or visible light can be used to excite
an electron from the ground valence states to the excited conduction states.
In a bulk semiconductor the band gap is fixed, and the energy states are
continuous. This results in a rather uniform absorption spectrum ((link]a).

2 is

350 0 450 500 ‘0 350 400
Wavelength [nm] Wavelength [nm]

UV-vis spectra of (a) bulk CdS and (b) 4 nm
CdS. Adapted from G. Kickelbick, Hybrid
Materials: Synthesis, Characterization and

Applications, Wiley-VCH, Weinheim
(2007).

In the case of Group 12-16 quantum dots, since the bandgap can be changed
with the size, these materials can absorb over a range of wavelengths. The
peaks seen in the absorption spectrum ({link]b) correspond to the optical
transitions between the electron and hole levels. The minimum energy and
thus the maximum wavelength peak corresponds to the first exciton peak or
the energy for an electron to get excited from the highest valence state to
the lowest conduction state. The quantum dot will not absorb wavelengths
of energy longer than this wavelength. This is known as the absorption
onset.

Fluorescence

Fluorescence is the emission of electromagnetic radiation in the form of
light by a material that has absorbed a photon. When a semiconductor
quantum dot (QD) absorbs a photon/energy equal to or greater than its band
gap, the electrons in the QD’s get excited to the conduction state. This
excited state is however not stable. The electron can relax back to its
ground state by either emitting a photon or lose energy via heat losses.
These processes can be divided into two categories — radiative decay and

non-radiative decay. Radiative decay is the loss of energy through the
emission of a photon or radiation. Non-radiative decay involves the loss of
heat through lattice vibrations and this usually occurs when the energy
difference between the levels is small. Non-radiative decay occurs much
faster than radiative decay.

Usually the electron relaxes to the ground state through a combination of
both radiative and non-radiative decays. The electron moves quickly
through the conduction energy levels through small non-radiative decays
and the final transition across the band gap is via a radiative decay. Large
nonradiative decays don’t occur across the band gap because the crystal
structure can’t withstand large vibrations without breaking the bonds of the
crystal. Since some of the energy is lost through the non-radiative decay, the
energy of the emitted photon, through the radiative decay, is much lesser
than the absorbed energy. As a result the wavelength of the emitted photon
or fluorescence is longer than the wavelength of absorbed light. This energy
difference is called the Stokes shift. Due this Stokes shift, the emission peak
corresponding to the absorption band edge peak is shifted towards a higher
wavelength (lower energy), i.e., [link].

Normalized U
Absorbance/PL intensity

400 500 600 700
Wavelength (nm)

Absorption spectra (a)
and emission spectra (b)
of CdSe tetrapod.

Intensity of emission versus wavelength is a bell-shaped Gaussian curve. As
long as the excitation wavelength is shorter than the absorption onset, the
maximum emission wavelength is independent of the excitation
wavelength. [link] shows a combined absorption and emission spectrum for
a typical CdSe tetrapod.

Factors affecting the optical properties of NPs

There are various factors that affect the absorption and emission spectra for
Group 12-16 semiconductor quantum crystals. Fluorescence is much more
sensitive to the background, environment, presence of traps and the surface
of the QDs than UV-visible absorption. Some of the major factors
influencing the optical properties of quantum nanoparticles include:

e Surface defects, imperfection of lattice, surface charges — The
surface defects and imperfections in the lattice structure of
semiconductor quantum dots occur in the form of unsatisfied
valencies. Similar to surface charges, unsatisfied valencies provide a
sink for the charge carriers, resulting in unwanted recombinations.

¢ Surface ligands — The presence of surface ligands is another factor
that affects the optical properties. If the surface ligand coverage is a
100%, there is a smaller chance of surface recombinations to occur.

¢ Solvent polarity — The polarity of solvents is very important for the
optical properties of the nanoparticles. If the quantum dots are
prepared in organic solvent and have an organic surface ligand, the
more non-polar the solvent, the particles are more dispersed. This
reduces the loss of electrons through recombinations again, since when
particles come in close proximity to each other, increases the non-
radiative decay events.

Applications of the optical properties of Group 12-16 semiconductor
NPs

The size dependent optical properties of NP’s have many applications from
biomedical applications to solar cell technology, from photocatalysis to
chemical sensing. Most of these applications use the following unique
properties.

For applications in the field of nanoelectronics, the sizes of the quantum
dots can be tuned to be comparable to the scattering lengths, reducing the
scattering rate and hence, the signal to noise ratio. For Group 12-16 QDs to
be used in the field of solar cells, the bandgap of the particles can be tuned
so as to form absorb energy over a large range of the solar spectrum,
resulting in more number of excitons and hence more electricity. Since the
nanoparticles are so small, most of the atoms are on the surface. Thus, the
surface to volume ratio is very large for the quantum dots. In addition to a
high surface to volume ratio, the Group 12-16 QDs respond to light energy.
Thus quantum dots have very good photocatalytic properties. Quantum dots
show fluorescence properties, and emit visible light when excited. This
property can be used for applications as biomarkers. These quantum dots
can be tagged to drugs to monitor the path of the drugs. Specially shaped
Group 12-16 nanoparticles such as hollow shells can be used as drug
delivery agents. Another use for the fluorescence properties of Group 12-16
semiconductor QDs is in color-changing paints, which can change colors
according to the light source used.

Bibliography

e M. J. Schulz, V. N. Shanov, and Y. Yun, Nanomedicine - Design of
Particles, Sensors, Motors, Implants, Robots, and Devices, Artech
House, London (2009).

e S. V. Gapoenko, Optical Properties of Semiconductor Nanocrystals,
Cambridge University Press, Cambridge (2003).

e T. Pradeep, Nano: The Essentials. Understanding Nanoscience and
Nanotechnology, Tata McGraw-Hill, New Delhi (2007).

e G. Schmid, Nanoparticles: From Theory to Application, Wiley-VCH,
Weinheim (2004).

e A. L.Rogach, Semiconductor Nanocrystal Quantum Dots. Synthesis,
Assembly, Spectroscopy and Applications, Springer Wien, New York

(2008).
e G. Kickelbick, Hybrid Materials: Synthesis, Characterization and
Applications, Wiley-VCH, Weinheim (2007).

Characterization of Group 12-16 (II-VI) Semiconductor Nanoparticles by
UV-visible Spectroscopy

Quantum dots (QDs) as a general term refer to nanocrystals of
semiconductor materials, in which the size of the particles are comparable
to the natural characteristic separation of an electron-hole pair, otherwise
known as the exciton Bohr radius of the material. When the size of the
semiconductor nanocrystal becomes this small, the electronic structure of
the crystal is governed by the laws of quantum physics. Very small Group
12-16 (II-VI) semiconductor nanoparticle quantum dots, in the order of 2 -
10 nm, exhibit significantly different optical and electronic properties from
their bulk counterparts. The characterization of size dependent optical
properties of Group 12-16 semiconductor particles provide a lot of
qualitative and quantitative information about them — size, quantum yield,
monodispersity, shape and presence of surface defects. A combination of
information from both the UV-visible absorption and fluorescence,
complete the analysis of the optical properties.

UV-visible absorbance spectroscopy

Absorption spectroscopy, in general, refers to characterization techniques
that measure the absorption of radiation by a material, as a function of the
wavelength. Depending on the source of light used, absorption spectroscopy
can be broadly divided into infrared and UV-visible spectroscopy. The band
gap of Group 12-16 semiconductors is in the UV-visible region. This means
the minimum energy required to excite an electron from the valence states
of the Group 12-16 semiconductor QDs to its conduction states, lies in the
UV-visible region. This is also a reason why most of the Group 12-16
semiconductor quantum dot solutions are colored.

This technique is complementary to fluorescence spectroscopy, in that UV-
visible spectroscopy measures electronic transitions from the ground state
to the excited state, whereas fluorescence deals with the transitions from the
excited state to the ground state. In order to characterize the optical
properties of a quantum dot, it is important to characterize the sample with
both these techniques

In quantum dots, due to the very small number of atoms, the addition or
removal of one atom to the molecule changes the electronic structure of the
quantum dot dramatically. Taking advantage of this property in Group 12-
16 semiconductor quantum dots, it is possible to change the band gap of the
material by just changing the size of the quantum dot. A quantum dot can
absorb energy in the form of light over a range of wavelengths, to excite an
electron from the ground state to its excited state. The minimum energy that
is required to excite an electron, is dependent on the band gap of the
quantum dot. Thus, by making accurate measurements of light absorption at
different wavelengths in the ultraviolet and visible spectrum, a correlation
can be made between the band gap and size of the quantum dot. Group 12-
16 semiconductor quantum dots are of particular interest, since their band
gap lies in the visible region of the solar spectrum.

The UV-visible absorbance spectroscopy is a characterization technique in
which the absorbance of the material is studied as a function of wavelength.
The visible region of the spectrum is in the wavelength range of 380 nm
(violet) to 740 nm (red) and the near ultraviolet region extends to
wavelengths of about 200 nm. The UV-visible spectrophotometer analyzes
over the wavelength range 200 — 900 nm.

When the Group 12-16 semiconductor nanocrystals are exposed to light
having an energy that matches a possible electronic transition as dictated by
laws of quantum physics, the light is absorbed and an exciton pair is
formed. The UV-visible spectrophotometer records the wavelength at which
the absorption occurs along with the intensity of the absorption at each
wavelength. This is recorded in a graph of absorbance of the nanocrystal
versus wavelength.

Instrumentation

A working schematic of the UV-visible spectrophotometer is shown in
[link].

Slit Rotating Disc Mirror

‘a | Q =] aN
| Sample cell

\
, an
Diffraction

grating

Light source Reference cell

‘X% mes >f---> a— Detector

Mirror Rotating Disc |

MJ

Chart Recorder

Schematic of UV-visible spectrophotometer.

The light source

Since it is a UV-vis spectrophotometer, the light source ([link]) needs to
cover the entire visible and the near ultra-violet region (200 - 900 nm).
Since it is not possible to get this range of wavelengths from a single lamp,
a combination of a deuterium lamp for the UV region of the spectrum and
tungsten or halogen lamp for the visible region is used. This output is then
sent through a diffraction grating as shown in the schematic.

The diffraction grating and the slit

The beam of light from the visible and/or UV light source is then separated

into its component wavelengths (like a very efficient prism) by a diffraction
grating ([link]). Following the slit is a slit that sends a monochromatic beam
into the next section of the spectrophotometer.

Rotating discs

Light from the slit then falls onto a rotating disc ([link]). Each disc consists
of different segments — an opaque black section, a transparent section and a
mirrored section. If the light hits the transparent section, it will go straight
through the sample cell, get reflected by a mirror, hits the mirrored section
of a second rotating disc, and then collected by the detector. Else if the light
hits the mirrored section, gets reflected by a mirror, passes through the
reference cell, hits the transparent section of a second rotating disc and then
collected by the detector. Finally if the light hits the black opaque section, it
is blocked and no light passes through the instrument, thus enabling the
system to make corrections for any current generated by the detector in the
absence of light.

Sample cell, reference cell and sample preparation

For liquid samples, a square cross section tube sealed at one end is used.
The choice of cuvette depends on the following factors:

¢ Type of solvent - For aqueous samples, specially designed rectangular
quartz, glass or plastic cuvettes are used. For organic samples glass
and quartz cuvettes are used.

¢ Excitation wavelength — Depending on the size and thus, bandgap of
the 12-16 semiconductor nanoparticles, different excitation
wavelengths of light are used. Depending on the excitation
wavelength, different materials are used

Cuvette Wavelength (nm)

Visible only glass 380 - 780

Visible only plastic 380 - 780
UV plastic 220 - 780
Quartz 200 - 900

Cuvette materials and their wavelengths.

e Cost — Plastic cuvettes are the least expensive and can be discarded
after use. Though quartz cuvettes have the maximum utility, they are
the most expensive, and need to reused. Generally, disposable plastic
cuvettes are used when speed is more important than high accuracy.

The best cuvettes need to be very clear and have no impurities that might
affect the spectroscopic reading. Defects on the cuvette such as scratches,
can scatter light and hence should be avoided. Some cuvettes are clear only
on two sides, and can be used in the UV-Visible spectrophotometer, but
cannot be used for fluorescence spectroscopy measurements. For Group 12-
16 semiconductor nanoparticles prepared in organic solvents, the quartz
cuvette is chosen.

In the sample cell the quantum dots are dispersed in a solvent, whereas in
the reference cell the pure solvent is taken. It is important that the sample be
very dilute (maximum first exciton absorbance should not exceed 1 au) and
the solvent is not UV-visible active. For these measurements, it is required
that the solvent does not have characteristic absorption or emission in the
region of interest. Solution phase experiments are preferred, though it is
possible to measure the spectra in the solid state also using thin films,
powders, etc. The instrumentation for solid state UV-visible absorption
spectroscopy is slightly different from the solution phase experiments and is
beyond the scope of discussion.

Detector

Detector converts the light into a current signal that is read by a computer.
Higher the current signal, greater is the intensity of the light. The computer

then calculates the absorbance using the in [link], where A denotes
absorbance, I is sample cell intensity and I, is the reference cell intensity.
Equation:

A = logio(ly/D

The following cases are possible:

Where I < Ip and A < 0. This usually occurs when the solvent absorbs
in the wavelength range. Preferably the solvent should be changed, to
get an accurate reading for actual reference cell intensity.

Where I = Ip and A= 0. This occurs when pure solvent is put in both
reference and sample cells. This test should always be done before
testing the sample, to check for the cleanliness of the cuvettes.

When A = 1. This occurs when 90% or the light at a particular
wavelength has been absorbed, which means that only 10% is seen at
the detector. So Ip/I becomes 100/10 = 10. Logo of 10 is 1.

When A > 1. This occurs in extreme case where more than 90% of the
light is absorbed.

Output

The output is the form of a plot of absorbance against wavelength, e.g.,
[link].

First exciton peak

Normalized UV
Absorbance/PL intensity

=
S

500 600 700
Wavelength (in nm)

Representative UV-visble
absorption spectrum for CdSe
tetrapods.

Beer-Lambert law

In order to make comparisons between different samples, it is important that
all the factors affecting absorbance should be constant except the sample
itself.

Effect of concentration on absorbance

The extent of absorption depends on the number of absorbing nanoparticles
or in other words the concentration of the sample. If it is a reasonably
concentrated solution, it will have a high absorbance since there are lots of
nanoparticles to interact with the light. Similarly in an extremely dilute
solution, the absorbance is very low. In order to compare two solutions, it is
important that we should make some allowance for the concentration.

Effect of container shape

Even if we had the same concentration of solutions, if we compare two
solutions — one in a rectagular shaped container (e.g., [link]) so that light
travelled 1 cm through it and the other in which the light travelled 100 cm
through it, the absorbance would be different. This is because if the length
the light travelled is greater, it means that the light interacted with more
number of nanocrystals, and thus has a higher absorbance. Again, in order
to compare two solutions, it is important that we should make some
allowance for the concentration.

icm

SA
|

|

A typical
rectangular
cuvette for
UV-visible

spectroscopy

The law

The Beer-Lambert law addresses the effect of concentration and container
shape as shown in [link], [link] and [link], where A denotes absorbance; € is
the molar absorptivity or molar absorption coefficient; | is the path length of
light (in cm); and c is the concentration of the solution (mol/dm?).
Equation:

logio(p/D = ele

Equation:

A = ésle

Molar absorptivity

From the Beer-Lambert law, the molar absorptivity 'e' can be expressed as
shown in [link].
Equation:

c = Alle

Molar absorptivity corrects for the variation in concentration and length of
the solution that the light passes through. It is the value of absorbance when
light passes through 1 cm of a 1 mol/dm? solution.

Limitations of Beer-Lambert law

The linearity of the Beer-Lambert law is limited by chemical and
instrumental factors.

e At high concentrations (> 0.01 M), the relation between absorptivity
coefficient and absorbance is no longer linear. This is due to the
electrostatic interactions between the quantum dots in close proximity.

e If the concentration of the solution is high, another effect that is seen is
the scattering of light from the large number of quantum dots.

e The spectrophotometer performs calculations assuming that the
refractive index of the solvent does not change significantly with the
presence of the quantum dots. This assumption only works at low
concentrations of the analyte (quantum dots).

e Presence of stray light.

Analysis of data

The data obtained from the spectrophotometer is a plot of absorbance as a
function of wavelength. Quantitative and qualitative data can be obtained
by analysing this information

Quantitative Information

The band gap of the semiconductor quantum dots can be tuned with the size
of the particles. The minimum energy for an electron to get excited from the
ground state is the energy to cross the band gap. In an absorption spectra,
this is given by the first exciton peak at the maximum wavelength (Ajax).

Size of the quantum dots

The size of quantum dots can be approximated corresponding to the first
exciton peak wavelength. Emperical relationships have been determined
relating the diameter of the quantum dot to the wavelength of the first
exciton peak. The Group 12-16 semiconductor quantum dots that they
studied were cadmium selenide (CdSe), cadmium telluride (CdTe) and
cadmium sulfide (CdS). The empirical relationships are determined by
fitting experimental data of absorbance versus wavelength of known sizes
of particles. The empirical equations determined are given for CdTe, CdSe,
and CdS in [link], [link] and [link] respectively, where D is the diameter
and A is the wavelength corresponding to the first exciton peak. For
example, if the first exciton peak of a CdSe quantum dot is 500 nm, the
corresponding diameter of the quantum dot is 2.345 nm and for a
wavelength of 609 nm, the corresponding diameter is 5.008 nm.
Equation:

D = (9.8127 x 10°7)A3 - (1.7147 x 10°3)A2 + (1.0064)d - 194.84

Equation:

D = (1.6122 x 10°7)A3 - (2.6575 x 10°)A2 + (1.6242 x 10°3)A + 41.57

Equation:

D = (-6.6521 x 10°8)h3 + (1.9577 x 10-)22 - (9.2352 x 102)h + 13.29

Concentration of sample

Using the Beer-Lambert law, it is possible to calculate the concentration of
the sample if the molar absorptivity for the sample is known. The molar
absorptivity can be calculated by recording the absorbance of a standard
solution of 1 mol/dm? concentration in a standard cuvette where the light
travels a constant distance of 1 cm. Once the molar absorptivity and the
absorbance of the sample are known, with the length the light travels being
fixed, it is possible to determine the concentration of the sample solution.

Empirical equations can be determined by fitting experimental data of
extinction coefficient per mole of Group 12-16 semiconductor quantum
dots, at 250 °C, to the diameter of the quantum dot, [link], [link], and [link].
Equation:

é = 10043 x D??
Equation:

€ = 5857 x D?*®
Equation:

é€ = 21536 x D*3

The concentration of the quantum dots can then be then be determined by
using the Beer Lambert law as given by [link].

Qualitative Information

Apart from quantitative data such as the size of the quantum dots and
concentration of the quantum dots, a lot of qualitative information can be
derived from the absorption spectra.

Size distribution

If there is a very narrow size distribution, the first exciton peak will be very
sharp ({link]). This is because due to the narrow size distribution, the
differences in band gap between different sized particles will be very small
and hence most of the electrons will get excited over a smaller range of
wavelengths. In addition, if there is a narrow size distribution, the higher
exciton peaks are also seen clearly.

(a)

Normalized UV
Absorbance/PL intensity
Normalized UV
Absorbance/PL intensity

-

400 500 600 700 400 500 600 700
Wavelength (nm) Wavelength (nm)

Narrow emission spectra (a) and broad
emission spectra (b) of CdSe QDs.

Shaped particles

In the case of a spherical quantum dot, in all dimensions, the particle is
quantum confined ([link]). In the case of a nanorod, whose length is not in
the quantum regime, the quantum effects are determined by the width of the

nanorod. Similar is the case in tetrapods or four legged structures. The
quantum effects are determined by the thickness of the arms. During the
synthesis of the shaped particles, the thickness of the rod or the arm of the
tetrapod does not vary among the different particles, as much as the length
of the rods or arms changes. Since the thickness of the rod or tetrapod is
responsible for the quantum effects, the absorption spectrum of rods and
tetrapods has sharper features as compared to a quantum dot. Hence,
qualitatively it is possible to differentiate between quantum dots and other
shaped particles.

Dot Rod Tetrapod

Different shaped nanoparticles
with the arrows indicating the
dimension where quantum
confinement effects are
observed.

Crystal lattice information

In the case of CdSe semiconductor quantum dots it has been shown that it is
possible to estimate the crystal lattice of the quantum dot from the
adsorption spectrum ([link]), and hence determine if the structure is zinc
blend or wurtzite.

- - -Zinc-Blende CdSe
~-™ —Wurtzite CdSe

Normalized UV Absorbance/
PLintensity

500 600 700
Wavelength (nm)

Zinc blende and wurtzite CdSe
absorption spectra. Adapted
from J. Jasieniak, C. Bullen, J.
van Embden, and P. Mulvaney,
J. Phys. Chem. B, 2005, 109,
20665.

UV-vis absorption spectra of Group 12-16 semiconductor
nanoparticles

Cadmium selenide

Cadmium selenide (CdSe) is one of the most popular Group 12-16
semiconductors. This is mainly because the band gap (712 nm or 1.74 eV)
energy of CdSe. Thus, the nanoparticles of CdSe can be engineered to have
a range of band gaps throughout the visible range, corresponding to the
major part of the energy that comes from the solar spectrum. This property
of CdSe along with its fluorescing properties is used in a variety of
applications such as solar cells and light emitting diodes. Though cadmium
and selenium are known carcinogens, the harmful biological effects of
CdSe can be overcome by coating the CdSe with a layer of zinc sulfide.
Thus CdSe, can also be used as bio-markers, drug-delivery agents, paints
and other applications.

A typical absorption spectrum of narrow size distribution wurtzite CdSe
quantum dot is shown in [link]. A size evolving absorption spectra is shown
in [link]. However, a complete analysis of the sample is possible only by
also studying the fluorescence properties of CdSe.

Absorbance (a.u.)

400 500 600 700 800
Wavelength (nm)

Wurtzite CdSe quantum dot.

Adapted from X. Zhong, Y.

Feng, and Y. Zhang, J. Phys.
Chem. C, 2007, 111, 526.

Absorbance (a.u.)

Wavelength (nm)

Size evolving absorption
spectra of CdSe quantum dots.

Cadmium telluride (CdTe)

Cadmium telluride has a band gap of 1.44 eV (860 nm) and as such it
absorbs in the infrared region. Like CdSe, the sizes of CdTe can be
engineered to have different band edges and thus, different absorption
spectra as a function of wavelength. A typical CdTe spectra is shown in
[link]. Due to the small bandgap energy of CdTe, it can be used in tandem
with CdSe to absorb in a greater part of the solar spectrum.

Absorbance (a.u.)

400 500 600 700
Wavelength (nm)

Size evolving absorption
spectra of CdTe quantum dots
from 3 nm to 7 nm. Adapted
from C. Qi-Fan, W. Wen-Xing,
G. Ying-Xin, L. Meng-Ying, X.
Shu-Kun and Z. Xiu-Juan,
Chin. J. Anal. Chem., 2007, 35,
135:

Other Group 12-16 semiconductor systems

[link] shows the bulk band gap of other Group 12-16 semiconductor
systems. The band gap of ZnS falls in the UV region, while those of ZnSe,
CdS, and ZnTe fall in the visible region.

Material Band gap (eV) Wavelength (nm)
ZnS 3.61 343.2
ZnSe 2.69 460.5
ZnTe 2.39 518.4
CdS 2.49 497.5
CdSe 1.74 712.1
CdTe 1.44 860.3

Bulk band gaps of different Group 12-16 semiconductors.

Heterostructures of Group 12-16 semiconductor systems

It is often desirable to have a combination of two Group 12-16
semiconductor system quantum heterostructures of different shapes like
dots and tetrapods, for applications in solar cells, bio-markers, etc. Some of
the most interesting systems are ZnS shell-CdSe core systems, such as the
CdSe/CdS rods and tetrapods.

[link] shows a typical absorption spectra of CdSe-ZnS core-shell system.
This system is important because of the drastically improved fluorescence
properties because of the addition of a wide band gap ZnS shell than the
core CdSe. In addition with a ZnS shell, CdSe becomes bio-compatible.

Absorbance (a.u.)

a ZnS layer 4

ZnS layer 1
CdSe core

0 4) 5D HO CO EH 70
Wavelength (nm)

Absorption spectra of CdSe
core, ZnS shell. Adapted from
C. Qing-Zhu, P. Wang, X. Wang
and Y. Li, Nanoscale Res. Lett.,
2008, 3, 213.

A CdSe seed, CdS arm nanorods system is also interesting. Combining
CdSe and CdS in a single nanostructure creates a material with a mixed
dimensionality where holes are confined to CdSe while electrons can move
freely between CdSe and CdS phases.

Bibliography

e S. V. Gapoenko, Optical Properties of Semiconductor Nanocrystals,
Cambridge University Press, Cambridge (2003).
e W. W. Yu, L. Qu, W. Guo, and X. Peng, Chem. Mater., 2003, 15, 2854.

J. Jasieniak, C. Bullen, J. van Embden, and P. Mulvaney, J. Phys.
Chem. B, 2005, 109, 20665.

X. Zhong, Y. Feng, and Y. Zhang, J. Phys. Chem. C, 2007, 111, 526.
D. V. Talapin, J. H. Nelson, E. V. Shevchenko, S. Aloni, B. Sadtler,
and A. P. Alivisatos, Nano Lett., 2007, 7, 2951.

C. Qing-Zhu, P. Wang, X. Wang, and Y. Li, Nanoscale Res. Lett.,
2008, 3, 213.

C. Qi-Fan, W. Wen-Xing, G. Ying-Xin, L. Meng-Ying, X. Shu-Kun,
and Z. Xiu-Juan, Chin. J. Anal. Chem., 2007, 35, 135.

Optical Characterization of Group 12-16 (II-VI) Semiconductor
Nanoparticles by Fluorescence Spectroscopy

Group 12-16 semiconductor nanocrystals when exposed to light of a
particular energy absorb light to excite electrons from the ground state to
the excited state, resulting in the formation of an electron-hole pair (also
known as excitons). The excited electrons relax back to the ground state,
mainly through radiative emission of energy in the form of photons.

Quantum dots (QD) refer to nanocrystals of semiconductor materials where
the size of the particles is comparable to the natural characteristic separation
of an electron-hole pair, otherwise known as the exciton Bohr radius of the
material. In quantum dots, the phenomenon of emission of photons
associated with the transition of electrons from the excited state to the
ground state is called fluorescence.

Fluorescence spectroscopy

Emission spectroscopy, in general, refers to a characterization technique
that measures the emission of radiation by a material that has been excited.
Fluorescence spectroscopy is one type of emission spectroscopy which
records the intensity of light radiated from the material as a function of
wavelength. It is a nondestructive characterization technique.

After an electron is excited from the ground state, it needs to relax back to
the ground state. This relaxation or loss of energy to return to the ground
state, can be achieved by a combination of non-radiative decay (loss of
energy through heat) and radiative decay (loss of energy through light).
Non-radiative decay by vibrational modes typically occurs between energy
levels that are close to each other. Radiative decay by the emission of light
occurs when the energy levels are far apart like in the case of the band gap.
This is because loss of energy through vibrational modes across the band
gap can result in breaking the bonds of the crystal. This phenomenon is
shown in [link].

Excited states

Nonradiative relaxation

Conduction band
Excitation
photon

t

Band gap N\I\VI>

Valence band

Emission of luminescence photon for Group 12-16
semiconductor quantum dot.

The band gap of Group 12-16 semiconductors is in the UV-visible region.
Thus, the wavelength of the emitted light as a result of radiative decay is
also in the visible region, resulting in fascinating fluorescence properties.

A fluorimeter is a device that records the fluorescence intensity as a
function of wavelength. The fluorescence quantum yield can then be
calculated by the ratio of photons absorbed to photons emitted by the
system. The quantum yield gives the probability of the excited state getting
relaxed via fluorescence rather than by any other non-radiative decay.

Difference between fluorescence and phosphorescence

Photoluminescence is the emission of light from any material due to the
loss of energy from excited state to ground state. There are two main types
of luminescence — fluorescence and phosphorescence. Fluorescence is a fast
decay process, where the emission rate is around 10° s"! and the lifetime is
around 1079 - 10°” s. Fluorescence occurs when the excited state electron
has an opposite spin compared to the ground state electrons. From the laws
of quantum mechanics, this is an allowed transition, and occurs rapidly by
emission of a photon. Fluorescence disappears as soon as the exciting light
source is removed.

Phosphorescence is the emission of light, in which the excited state electron
has the same spin orientation as the ground state electron. This transition is
a forbidden one and hence the emission rates are slow (10? - 10° s“!). So the
phosphorescence lifetimes are longer, typically seconds to several minutes,
while the excited phosphors slowly returned to the ground state.
Phosphorescence is still seen, even after the exciting light source is
removed. Group 12-16 semiconductor quantum dots exhibit fluorescence
properties when excited with ultraviolet light.

Instrumentation

The working schematic for the fluorometer is shown in [link].

Primary filter
\
| Sample cell
\
\

Wavelengths

. tedb

n0Ns Diffraction crea y
fluorescent

Grating
; compound plus
Light source stray light
Secondary filter ——_ <=
Wavelengths specific to

compound

Detector

Output

Schematic of fluorometer.

The light source

The excitation energy is provided by a light source that can emit
wavelengths of light over the ultraviolet and the visible range. Different
light sources can be used as excitation sources such as lasers, xenon arcs
and mercury-vapor lamps. The choice of the light source depends on the
sample. A laser source emits light of a high irradiance at a very narrow
wavelength interval. This makes the need for the filter unnecessary, but the
wavelength of the laser cannot be altered significantly. The mercury vapor
lamp is a discrete line source. The xenon arc has a continuous emission
spectrum between the ranges of 300 - 800 nm.

The diffraction grating and primary filter

The diffraction grating splits the incoming light source into its component
wavelengths ([link]). The monochromator can then be adjusted to choose
with wavelengths to pass through. Following the primary filter, specific
wavelengths of light are irradiated onto the sample

Sample cell and sample preparation

A proportion of the light from the primary filter is absorbed by the sample.
After the sample gets excited, the fluorescent substance returns to the
ground state, by emitting a longer wavelength of light in all directions
({link]). Some of this light passes through a secondary filter. For liquid
samples, a square cross section tube sealed at one end and all four sides
clear, is used as a sample cell. The choice of cuvette depends on three
factors:

1. Type of solvent - For aqueous samples, specially designed rectangular
quartz, glass or plastic cuvettes are used. For organic samples glass
and quartz cuvettes are used.

2. Excitation wavelength — Depending on the size and thus, bandgap of
the Group 12-16 semiconductor nanoparticles, different excitation
wavelengths of light are used. Depending on the excitation
wavelength, different materials are used ([link]).

Cuvette Wavelength (nm)

Visible only glass 380 - 780
Visible only plastic 380 - 780
UV plastic 220 - 780
Quartz 200 - 900

Cuvette materials and their wavelengths.

3. Cost — Plastic cuvettes are the least expensive and can be discarded
after use. Though quartz cuvettes have the maximum utility, they are
the most expensive, and need to reused. Generally, disposable plastic
cuvettes are used when speed is more important than high accuracy.

icm
_

oN; :

%
|
|

A typical
cuvette for
fluorescence
spectroscopy

The cuvettes have a 1 cm path length for the light ((link]). The best cuvettes
need to be very clear and have no impurities that might affect the

spectroscopic reading. Defects on the cuvette, such as scratches, can scatter
light and hence should be avoided. Since the specifications of a cuvette are
the same for both, the UV-visible spectrophotometer and fluorimeter, the
Same cuvette that is used to measure absorbance can be used to measure the
fluorescence. For Group 12-16 semiconductor nanoparticles preparted in
organic solvents, the clear four sided quartz cuvette is used. The sample
solution should be dilute (absorbance <1 au), to avoid very high signal from
the sample to burn out the detector. The solvent used to disperse the
nanoparticles should not absorb at the excitation wavelength.

Secondary filter

The secondary filter is placed at a 90° angle ([link]) to the original light
path to minimize the risk of transmitted or reflected incident light reaching
the detector. Also this minimizes the amount of stray light, and results in a
better signal-to-noise ratio. From the secondary filter, wavelengths specific
to the sample are passed onto the detector.

Detector

The detector can either be single-channeled or multichanneled ([link]). The
single-channeled detector can only detect the intensity of one wavelength at
a time, while the multichanneled detects the intensity at all wavelengths
simultaneously, making the emission monochromator or filter unnecessary.
The different types of detectors have both advantages and disadvantages.

Output

The output is the form of a plot of intensity of emitted light as a function of
wavelength as shown in [link].

Photoluminescence
intensity

500 550 600 650 700
Wavelength (nm)

Emission spectra of CdSe
quantum dot.

Analysis of data

The data obtained from fluorimeter is a plot of fluorescence intensity as a
function of wavelength. Quantitative and qualitative data can be obtained
by analysing this information.

Quantitative information

From the fluorescence intensity versus wavelength data, the quantum yield
(®,) of the sample can be determined. Quantum yield is a measure of the
ratio of the photons absorbed with respect to the photons emitted. It is
important for the application of Group 12-16 semiconductor quantum dots
using their fluorescence properties, for e.g., bio-markers.

The most well-known method for recording quantum yield is the
comparative method which involves the use of well characterized standard
solutions. If a test sample and a standard sample have similar absorbance
values at the same excitation wavelength, it can be assumed that the number
of photons being absorbed by both the samples is the same. This means that
a ratio of the integrated fluorescence intensities of the test and standard

sample measured at the same excitation wavelength will give a ratio of
quantum yields. Since the quantum yield of the standard solution is known,
the quantum yield for the unknown sample can be calculated.

A plot of integrated fluorescence intensity versus absorbance at the
excitation wavelength is shown in [link]. The slope of the graphs shown in
[link] are proportional to the quantum yield of the different samples.
Quantum yield is then calculated using [link], where subscripts ST denotes
standard sample and X denotes the test sample; QY is the quantum yield; RI
is the refractive index of the solvent.

100 + @ Standardsample
so — m Test sample

Integrated fluorescence
intensity

0+ : -

0 0.05 0.1
Wavelength (nm)

Integrated fluoresncene intensity as a function of
absorbance.

Equation:

QY, = slopey (RIx)*
OY sr

slopecy (Rl gq)?

Take the example of [link]. If the same solvent is used in both the sample
and the standard solution, the ratio of quantum yields of the sample to the

standard is given by [link]. If the quantum yield of the standard is known to
0.95, then the quantum yield of the test sample is 0.523 or 52.3%.
Equation:
OY = Tat
OY, 2.56

The assumption used in the comparative method is valid only in the Beer-
Lambert law linear regime. Beer-Lambert law states that absorbance is
directly proportional to the path length of light travelled within the sample,
and concentration of the sample. The factors that affect the quantum yield
measurements are the following:

e Concentration — Low concentrations should be used (absorbance <
0.2 a.u.) to avoid effects such as self quenching.

e Solvent — It is important to take into account the solvents used for the
test and standard solutions. If the solvents used for both are the same
then the comparison is trivial. However, if the solvents in the test and
standard solutions are different, this difference needs to be accounted
for. This is done by incorporating the solvent refractive indices in the
ratio calculation.

e Standard samples — The standard samples should be characterized
thoroughly. In addition, the standard sample used should absorb at the
excitation wavelength of the test sample.

e Sample preparation — It is important that the cuvettes used are clean,
scratch free and clear on all four sides. The solvents used must be of
spectroscopic grade and should not absorb in the wavelength range.

e Slit width — The slit widths for all measurements must be kept
constant.

The quantum yield of the Group 12-16 semiconductor nanoparticles are
affected by many factors such as the following.

e Surface defects — The surface defects of semiconductor quantum dots
occur in the form of unsatisfied valencies. Thus resulting in unwanted
recombinations. These unwanted recombinations reduce the loss of
energy through radiative decay, and thus reducing the fluorescence.

¢ Surface ligands — If the surface ligand coverage is a 100%, there is a
smaller chance of surface recombinations to occur.

¢ Solvent polarity — If the solvent and the ligand have similar solvent
polarities, the nanoparticles are more dispersed, reducing the loss of
electrons through recombinations.

Qualitative Information

Apart from quantum yield information, the relationship between intensity of
fluorescence emission and wavelength, other useful qualitative information
such as size distribution, shape of the particle and presence of surface
defects can be obtained.

As shown in [link], the shape of the plot of intensity versus wavelength is a
Gaussian distribution. In [link], the full width at half maximum (FWHM) is
given by the difference between the two extreme values of the wavelength
at which the photoluminescence intensity is equal to half its maximum
value. From the full width half max (FWHM) of the fluorescence intensity
Gaussian distribution, it is possible to determine qualitatively the size
distribution of the sample. For a Group 12-16 quantum dot sample if the
FWHM is greater than 30, the system is very polydisperse and has a large
size distribution. It is desirable for all practical applications for the FWHM
to be lesser than 30.

FWHM
Max
intensity

% x Max
intensity

Photoluminescence
intensity (a.u.)

500 550 600 650 700
Wavelength (nm)

Emission spectra of CdSe QDs
showing the full width half maximum
(FWHM).

From the FWHM of the emission spectra, it is also possible to qualitatively
get an idea if the particles are spherical or shaped. During the synthesis of
the shaped particles, the thickness of the rod or the arm of the tetrapod does
not vary among the different particles, as much as the length of the rods or
arms changes. The thickness of the arm or rod is responsible for the
quantum effects in shaped particles. In the case of quantum dots, the
particle is quantum confined in all dimensions. Thus, any size distribution
during the synthesis of quantum dots greatly affects the emission spectra.
As a result the FWHM of rods and tetrapods is much smaller as compared
to a quantum dot. Hence, qualitatively it is possible to differentiate between
quantum dots and other shaped particles.

Another indication of branched structures is the decrease in the intensity of
fluorescence peaks. Quantum dots have very high fluorescence values as
compared to branched particles, since they are quantum confined in all
dimensions as compared to just 1 or 2 dimensions in the case of branched
particles.

Fluorescence spectra of different Group 12-16 semiconductor
nanoparticles

The emission spectra of all Group 12-16 semiconductor nanoparticles are
Gaussian curves as shown in [link] and [link]. The only difference between
them is the band gap energy, and hence each of the Group 12-16
semiconductor nanoparticles fluoresce over different ranges of wavelengths

Cadmium selenide

Since its bulk band gap (1.74 eV, 712 nm) falls in the visible region
cadmium Selenide (CdSe) is used in various applications such as solar cells,
light emitting diodes, etc. Size evolving emission spectra of cadmium
selenide is shown in [link]. Different sized CdSe particles have different
colored fluorescence spectra. Since cadmium and selenide are known
carcinogens and being nanoparticles are easily absorbed into the human
body, there is some concern regarding these particles. However, CdSe
coated with ZnS can overcome all the harmful biological effects, making
cadmium selenide nanoparticles one of the most popular 12-16
semiconductor nanoparticle.

Photoluminescence
intensity
(arb units)

450 500 550 £600
Wavelength (nm)

Size evolving CdSe emission
spectra. Adapted from
http://www. physics.mq.edu.au.

A combination of the absorbance and emission spectra is shown in [link]
for four different sized particles emitting green, yellow, orange, and red
fluorescence.

4
n # ® t 4 x f 0 % A
i ee fos
a a * a “ea
»* »>* ¥ ‘ \

ty
.

Absorbance/PLintensi

450 500 550 600 650 700

Wavelength (in nm)

Absorption and emission spectra of CdSe quantum
dots. Adapted from G. Schmid, Nanoparticles:
From Theory to Application, Wiley-VCH,
Weinham (2004).

Cadmium telluride

Cadmium Telluride (CdTe) has a band gap of 1.44 eV and thus absorbs in
the infra red region. The size evolving CdTe emission spectra is shown in
[link].

Photoluminescence
intensity

Wavelength (nm)

Size evolution spectra of CdTe
quantum dots.

Adding shells to QDs

Capping a core quantum dot with a semiconductor material with a wider
bandgap than the core, reduces the nonradiative recombination and results
in brighter fluorescence emission. Quantum yields are affected by the
presences of free surface charges, surface defects and crystal defects, which
results in unwanted recombinations. The addition of a shell reduces the
nonradiative transitions and majority of the electrons relax radiatively to the
valence band. In addition, the shell also overcomes some of the surface
defects.

For the CdSe-core/ZnS-shell systems exhibit much higher quantum yield as
compared to core CdSe quantum dots as seen in [link].

CdSe core/ZnS shell

/\

/\

Photoluminescence
intensity
oB8BBSSBSBSSBEBB

j \
/ \
/ CoreCdSe \
\
N\
§00 550 600
Wavelength (nm)

Emission spectra of
core CdSe only and
CdSe-core/ZnS-
shell.

Bibliography

e A. T.R. Williams, S. A. Winfield, and J. N. Miller, Analyst, 1983, 108,
1067.

e G. Schmid, Nanoparticles: From Theory to Application, Wiley-VCH,
Weinham, (2004).

e J. Y. Hariba, A Guide to Recording Fluorescence Quantum Yield, Jobin
Yvon Hariba Limited, Stanmore (2003).

e C. Qing Zhu, P. Wang, X. Wang, and Y. Li, Nanoscale Res. Lett..,
2008, 3, 213.

Carbon Nanomaterials

Introduction

Although nanomaterials had been known for many years prior to the report
of Cgp the field of nanoscale science was undoubtedly founded upon this
seminal discovery. Part of the reason for this explosion in nanochemistry is
that while carbon materials range from well-defined nano sized molecules
(i.e., Cgq) to tubes with lengths of hundreds of microns, they do not exhibit
the instabilities of other nanomaterials as a result of the very high activation
barriers to their structural rearrangement. As a consequence they are highly
stable even in their unfunctionalized forms. Despite this range of carbon
nanomaterials possible they exhibit common reaction chemistry: that of
organic chemistry.

The previously unknown allotrope of carbon: Cgg, was discovered in 1985,
and in 1996, Curl, Kroto, and Smalley were awarded the Nobel Prize in
Chemistry for the discovery. The other allotropes of carbon are graphite
(sp*) and diamond (sp?). Cg, commonly known as the “buckyball” or
“Buckminsterfullerene”, has a spherical shape comprising of highly
pyramidalized sp* carbon atoms. The Ceo variant is often compared to the
typical soccer football, hence buckyball. However, confusingly, this term is
commonly used for higher derivatives. Fullerenes are similar in sheet
structure to graphite but they contain pentagonal (or sometimes heptagonal)
rings that prevent the sheet from being planar. The unusual structure of Cgg
led to the introduction of a new class of molecules known as fullerenes,
which now constitute the third allotrope of carbon. Fullerenes are
commonly defined as “any of a class of closed hollow aromatic carbon
compounds that are made up of twelve pentagonal and differing numbers of
hexagonal faces.”

The number of carbon atoms in a fullerene range from Cgp to C79, C76, and
higher. Higher order fullerenes include carbon nanotubes that can be
described as fullerenes that have been stretched along a rotational axis to
form a tube. As a consequence of differences in the chemistry of fullerenes
such as Cgg and C79 as compared to nanotubes, these will be dealt with
separately herein. In addition there have also been reports of nanohorns and

nanofibers, however, these may be considered as variations on the general
theme. It should be noted that fullerenes and nanotubes have been shown to
be in flames produced by hydrocarbon combustion. Unfortunately, these
naturally occurring varieties can be highly irregular in size and quality, as
well as being formed in mixtures, making them unsuitable for both research
and industrial applications.

Fullerenes

Carbon-60 (Cgp) is probably the most studied individual type of
nanomaterial. The spherical shape of Cgg is constructed from twelve
pentagons and twenty hexagons and resembles a soccer ball ([link]a). The
next stable higher fullerene is C7p ({link]b) that is shaped like a rugby or
American football. The progression of higher fullerenes continues in the
sequence C74, C76, C7, etc. The structural relationship between each
involves the addition of six membered rings. Mathematically (and
chemically) two principles define the existence of a stable fullerene, i.e.,
Euler’s theorem and isolated pentagon rule (IPR). Euler’s theorem states
that for the closure of each spherical network, n (n = 2) hexagons and 12
pentagons are required while the IPR says no two pentagons may be
connected directly with each other as destabilization is caused by two
adjacent pentagons.

(2) (b)

Molecular structures of (a) Cgp and (b) C7.

Although fullerenes are composed of sp? carbons in a similar manner to
graphite, fullerenes are soluble in various common organic solvents. Due to
their hydrophobic nature, fullerenes are most soluble in CS (Cgg = 7.9
mg/mL) and toluene (Cgp = 2.8 mg/mL). Although fullerenes have a
conjugated system, their aromaticity is distinctive from benzene that has all
C-C bonds of equal lengths, in fullerenes two distinct classes of bonds exist.
The shorter bonds are at the junctions of two hexagons ([6, 6] bonds) and
the longer bonds at the junctions of a hexagon and a pentagon ([5,6] bonds).
This difference in bonding is responsible for some of the observed
reactivity of fullerenes.

Synthesis of fullerenes

The first observation of fullerenes was in molecular beam experiments at
Rice University. Subsequent studies demonstrated that Cg¢p it was relatively
easy to produce grams of fullerenes. Although the synthesis is relatively
straightforward fullerene purification remains a challenge and determines
fullerene’s commercial price. The first method of production of measurable
quantities of fullerenes used laser vaporization of carbon in an inert
atmosphere, but this produced microscopic amounts of fullerenes.
Laboratory scales of fullerene are prepared by the vaporization of carbon
rods in a helium atmosphere. Commercial production ordinarily employs a
simple ac or dc arc. The fullerenes in the black soot collected are extracted
in toluene and purified by liquid chromatography. The magenta Cgg comes
off the column first, followed by the red Cyo, and other higher fullerenes.
Even though the mechanism of a carbon arc differs from that of a resistively
heated carbon rod (because it involves a plasma) the He pressure for
optimum Cg formation is very similar.

A ratio between the mass of fullerenes and the total mass of carbon soot
defines fullerene yield. The yields determined by UV-Vis absorption are
approximately 40%, 10-15%, and 15% in laser, electric arc, and solar
processes. Interestingly, the laser ablation technique has both the highest
yield and the lowest productivity and, therefore, a scale-up to a higher

power is costly. Thus, fullerene commercial production is a challenging
task. The world's first computer controlled fullerene production plant is now
operational at the MER Corporation, who pioneered the first commercial
production of fullerene and fullerene products.

Endohedral fullerenes

Endohedral fullerenes are fullerenes that have incorporated in their inner
sphere atoms, ions or clusters. Endohedral fullerenes are generally divided
into two groups: endohedral metallofullerenes and non-metal doped
fullerenes. The first endohedral metallofullerenes was called La@Cgo. The
@ sign in the name reflects the notion of a small molecule trapped inside a
shell.

Doping fullerenes with metals takes place in-situ during the fullerene
synthesis in an arc reactor or via laser evaporation. A wide range of metals
have been encased inside a fullerene, i.e., Sc, Y, La, Ce, Ba, Sr, K, U, Zr,
and Hf. Unfortunately, the synthesis of endohedral metallofullerenes is
unspecific because in addition a high yield of unfilled fullerenes,
compounds with different cage sizes are prepared (e.g., La@Cgo or
La@Cg>). A characteristic of endohedral metallofullerenes is that electrons
will transfer from the metal atom to the fullerene cage and that the metal
atom takes a position off-center in the cage. The size of the charge transfer
is not always simple to determine, but it is usually between 2 and 3 units
(e.g., Lay>@Cgo) but can be as high as 6 electrons (e.g., ScsN@Cgo). These
anionic fullerene cages are very stable molecules and do not have the
reactivity associated with ordinary empty fullerenes (see below). This lack
of reactivity is utilized in a method to purify endohedral metallofullerenes
from empty fullerenes.

The endohedral He@Cgy and Ne@Cgp form when Ceo is exposed to a
pressure of around 3 bar of the appropriate noble gases. Under these
conditions it was possible to dope 1 in every 650,000 Cgg cages with a
helium atom. Endohedral complexes with He, Ne, Ar, Kr and Xe as well as
numerous adducts of the He@Cgy compound have also been proven with
operating pressures of 3000 bars and incorporation of up to 0.1 % of the

noble gases. The isolation of N@Cgop, N@C7p and P@Cegp is very unusual
and unlike the metal derivatives no charge transfer of the pnictide atom in
the center to the carbon atoms of the cage takes place.

Chemically functionalized fullerenes

Although fullerenes have a conjugated aromatic system all the carbons are
quatemary (i.e., containing no hydrogen), which results in making many of
the characteristic substitution reactions of planar aromatics impossible.
Thus, only two types of chemical transformations exist: redox reactions and
addition reactions. Of these, addition reactions have the largest synthetic
value. Another remarkable feature of fullerene addition chemistry is the
thermodymics of the process. Since the sp? carbon atoms in a fullerene are
paramidalized there is significant strain energy. For example, the strain
energy in Cg is ca 8 kcal/mol, which is 80% of its heat of formation. So the
relief of this strain energy leading to sp? hybridized C atoms is the major
driving force for addition reactions ([link]). As a consequence, most
additions to fullerenes are exothermic reactions.

seat) 101.6" A] JL y110.3"
' —" nee

Ceo (Sp) Cgo-adduct (sp)

Strain release after

addition of reagent

A to a pyramidalize
carbon of Cgo.

Cyclic voltammetry (CV) studies show that Cgg can be reduced and
oxidized reversibly up to 6 electrons with one-electron transfer processes.
Fulleride anions can be generated by electrochemical method and then be

used to synthesize covalent organofullerene derivatives. Alkali metals can
chemically reduce fullerene in solution and solid state to form M,Cgg (x = 3
- 6). Cg can also be reduced by less electropositive metals like mercury to
form Cg” and Cg’. In addition, salts can also be synthesized with organic
molecules, for example [TDAE"*][Cgq_] possesses interesting electronic and
magnetic behavior.

Geometric and electronic analysis predicted that fullerene behaves live an
electro-poor conjugated polyolefin. Indeed Cgp and C79 undergo a range of
nucleophilic reactions with carbon, nitrogen, phosphorous and oxygen
nucleophiles. C60 reacts readily with organolithium and Grignard
compounds to form alkyl, phenyl or alkanyl fullerenes. Possibly the most
widely used additions to fullerene is the Bingel reaction ([link]), where a
carbon nucleophile, generated by deprotonation of a-halo malonate esters or
ketones, is added to form a cyclopropanation product. The a-halo esters and
ketones can also be generated in situ with Ip or CBr, and a weak base as
1,8-diazabicyclo[5.4.0]unde-7ene (DBU). The Bingel reaction is considered
one of the most versatile and efficient methods to functionalize Cgpo.

EtO(OyC C(O)ORt

(O\OEt
Br + NaH

CKOORt
er

-H, - NaBr

Bingel reaction of Cgg with 2-
bromoethylmalonate.

Cycloaddition is another powerful tool to functionalize fullerenes, in
particular because of its selectivity with the 6,6 bonds, limiting the possible
isomers ([link]). The dienophilic feature of the [6,6] double bonds of Cgp
enables the molecule to undergo various cycloaddition reactions in which

the monoadducts can be generated in high yields. The best studies
cycloadditon reactions of fullerene are [3+2] additions with
diazoderivatives and azomethine ylides (Prato reactions). In this reaction,
azomethine ylides can be generated in situ from condensation of a-amino
acids with aldehydes or ketones, which produce 1,3 dipoles to further react
with Cgp in good yields ({link]). Hundreds of useful building blocks have
been generated by those two methods. The Prato reactions have also been
successfully applied to carbon nanotubes.

Geometrical shapes built onto a [6,6] ring
junction: a) open, b) four-membered ring, c)
five-membered ring, and d) six-membered ring.

CH;

Prato reaction of Cgq with N-methyglycine and
paraformaldehyde.

The oxidation of fullerenes, such as Cgo, has been of increasing interest
with regard to applications in photoelectric devices, biological systems, and
possible remediation of fullerenes. The oxidation of Cgg to CggO, (n = 1, 2)
may be accomplished by photooxidation, ozonolysis, and epoxidation. With
each of these methods, there is a limit to the isolable oxygenated product,
CeO, with n < 3. Highly oxygenated fullerenes, Cg gO, with 3 <n < 9, have
been prepared by the catalytic oxidation of Cgg with REMeO3/H20>.

Carbon nanotubes

A key breakthrough in carbon nanochemistry came in 1993 with the report
of needle-like tubes made exclusively of carbon. This material became
known as carbon nanotubes (CNTs). There are several types of nanotubes.
The first discovery was of multi walled tubes (MWNTs) resembling many
pipes nested within each other. Shortly after MWNTs were discovered
single walled nanotubes (SWNTs) were observed. Single walled tubes
resemble a single pipe that is potentially capped at each end. The properties
of single walled and multi walled tubes are generally the same, although
single walled tubes are believed to have superior mechanical strength and
thermal and electrical conductivity; it is also more difficult to manufacture
them.

Single walled carbon nanotubes (SWNTs) are by definition fullerene
materials. Their structure consists of a graphene sheet rolled into a tube and
capped by half a fullerene ([link]). The carbon atoms in a SWNT, like those
in a fullerene, are sp2 hybridized. The structure of a nanotube is analogous
to taking this graphene sheet and rolling it into a seamless cylinder. The
different types of SWNTs are defined by their diameter and chirality. Most
of the presently used single-wall carbon nanotubes have been synthesized
by the pulsed laser vaporization method, however, increasingly SWNTs are
prepared by vapor liquid solid catalyzed growth.

(b)

(c)

Structure of single walled carbon nanotubes
(SWNTs) with (a) armchair, (b) zig-zag, and (c)
chiral chirality.

The physical properties of SWNTs have made them an extremely attractive
material for the manufacturing of nano devices. SWNTs have been shown
to be stronger than steel as estimates for the Young’s modulus approaches 1
Tpa. Their electrical conductance is comparable to copper with anticipate
current densities of up to 10'° A/cm? and a resistivity as low as 0.34 x 104
Q.cm at room temperatures. Finally, they have a high thermal conductivity
(3000 - 6000 W.m/K).

The electronic properties of a particular SWNT structure are based on its
chirality or twist in the structure of the tube which is defined by its n,m
value. The values of n and m determine the chirality, or "twist" of the
nanotube. The chirality in turn affects the conductance of the nanotube, its
density, its lattice structure, and other properties. A SWNT is considered
metallic if the value n-m is divisible by three. Otherwise, the nanotube is
semi-conducting. The external environment also has an effect on the

conductance of a tube, thus molecules such as O» and NH3 can change the
overall conductance of a tube, while the presence of metals have been
shown to significantly effect the opto-electronic properties of SWNTs.

Multi walled carbon nanotubes (MWNTs) range from double walled NTs,
through many-walled NTs ([{link]) to carbon nanofibers. Carbon nanofibers
are the extreme of multi walled tubes ([link]) and they are thicker and
longer than either SWNTs or MWNTs, having a cross-sectional of ca. 500
A? and are between 10 to 100 pm in length. They have been used
extensively in the construction of high strength composites.

TEM image of an individual
multi walled carbon nanotube
(MWNTs). Copyright of
Nanotech Innovations.

A

“

Magn Det WD -}—-—————-{ 2um

Acc.V Spo e
30.0kV 3.0 15000x SE 18.6 Hivac

SEM image of vapor grown
carbon nanofibers.

Synthesis of carbon nanotubes

A range of methodologies have been developed to produce nanotubes in
sizeable quantities, including arc discharge, laser ablation, high pressure
carbon monoxide (HiPco), and vapor liquid solid (VLS) growth. All these
processes take place in vacuum or at low pressure with a process gases,
although VLS growth can take place at atmospheric pressure. Large
quantities of nanotubes can be synthesized by these methods; advances in
catalysis and continuous growth processes are making SWNTs more
commercially viable.

The first observation of nanotubes was in the carbon soot formed during the
arc discharge production of fullerenes. The high temperatures caused by the
discharge caused the carbon contained in the negative electrode to sublime
and the CNTs are deposited on the opposing electrode. Tubes produced by
this method were initially multi walled tubes (MWNTs). However, with the
addition of cobalt to the vaporized carbon, it is possible to grow single
walled nanotubes. This method it produces a mixture of components, and
requires further purification to separate the CNTs from the soot and the
residual catalytic metals. Producing CNTs in high yield depends on the

uniformity of the plasma arc, and the temperature of the deposit forming on
the carbon electrode.

Higher yield and purity of SWNTs may be prepared by the use of a dual-
pulsed laser. SWNTs can be grown in a 50% yield through direct
vaporization of a Co/Ni doped graphite rod with a high-powered laser in a
tube furnace operating at 1200 °C. The material produced by this method
appears as a mat of “ropes”, 10 - 20 nm in diameter and up to 100 pm or
more in length. Each rope consists of a bundle of SWNTs, aligned along a
common axis. By varying the process parameters such as catalyst
composition and the growth temperature, the average nanotube diameter
and size distribution can be varied. Although arc-discharge and laser
vaporization are currently the principal methods for obtaining small
quantities of high quality SWNTs, both methods suffer from drawbacks.
The first is that they involve evaporating the carbon source, making scale-
up on an industrial level difficult and energetically expensive. The second
issue relates to the fact that vaporization methods grow SWNTs in highly
tangled forms, mixed with unwanted forms of carbon and/or metal species.
The SWNTs thus produced are difficult to purify, manipulate, and assemble
for building nanotube-device architectures for practical applications.

In order to overcome some of the difficulties of these high-energy
processes, the chemical catalysis method was developed in which a
hydrocarbon feedstock is used in combination with a metal catalyst. The
catalyst is typically, but not limited to iron, colbalt, or iron/molybdenun,, it
is heated under reducing conditions in the presence of a suitable carbon
feedstock, e.g., ethylene. This method can be used for both SWNTs and
MWNTs; the formation of each is controlled by the identity of the catalyst
and the reaction conditions. A convenient laboratory scale apparatus is
available from Nanotech Innovations, Inc., for the synthesis of highly
uniform, consistent, research sample that uses pre-weighed catalyst/carbon
source ampoules. This system, allows for 200 mg samples of MWNTs to be
prepared for research and testing. The use of CO as a feedstock, in place of
a hydrocarbon, led to the development of the high-pressure carbon
monoxide (HiPco) procedure for SWNT synthesis. By this method, it is
possible to produce gram quantities of SWNTs, unfortunately, efforts to
scale beyond that have not met with complete success.

Initially developed for small-scale investigations of catalyst activity, vapor
liquid solid (VLS) growth of nanotubes has been highly studied, and now
shows promise for large-scale production of nanotubes. Recent approaches
have involved the use of well-defined nanoparticle or molecular precursors
and many different transition metals have been employed, but iron, nickel,
and cobalt remain to be the focus of most research. The nanotubes grow at
the sites of the metal catalyst; the carbon-containing gas is broken apart at
the surface of the catalyst particle, and the carbon is transported to the
edges of the particle, where it forms the nanotube. The length of the tube
grown in surface supported catalyst VLS systems appears to be dependent
on the orientation of the growing tube with the surface. By properly
adjusting the surface concentration and aggregation of the catalyst particles
it is possible to synthesize vertically aligned carbon nanotubes, i.e., as a
Carpet perpendicular to the substrate.

Of the various means for nanotube synthesis, the chemical processes show
the greatest promise for industrial scale deposition in terms of its price/unit
ratio. There are additional advantages to the VLS growth, which unlike the
other methods is capable of growing nanotubes directly on a desired
substrate. The growth sites are controllable by careful deposition of the
catalyst. Additionally, no other growth methods have been developed to
produce vertically aligned SWNTs.

Chemical functionalization of carbon nanotubes

The limitation of using carbon nanotubes in any practical applications has
been its solubility; for example SWNTs have little to no solubility in most
solvent due to the aggregation of the tubes. Aggregation/roping of
nanotubes occurs as a result of the high van der Waals binding energy of ca.
500 eV per mm of tube contact. The van der Waals force between the tubes
is so great, that it take tremendous energy to pry them apart, making it very
to make combination of nanotubes with other materials such as in
composite applications. The functionalization of nanotubes, i.e., the
attachment of “chemical functional groups” provides the path to overcome
these barriers. Functionalization can improve solubility as well as
processibility, and has been used to align the properties of nanotubes to

those of other materials. The clearest example of this is the ability to
solubilize nanotubes in a variety of solvents, including water. It is important
when discussing functionalization that a distinction is made between
covalent and non-covalent functionalization.

Current methods for solubilizing nanotubes without covalent
functionalization include highly aromatic solvents, super acids, polymers,
or surfactants. Non-covalent “functionalization” is generally on the concept
of supramolecular interactions between the SWNT and some
macromolecule as a result of various adsorption forces, such as van der
Waals’ and m-stacking interactions. The chemical speciation of the nanotube
itself is not altered as a result of the interaction. In contrast, covalent
functionalization relies on the chemical reaction at either the sidewall or
end of the SWNT. As may be expected the high aspect ratio of nanotubes
means that sidewall functionalization is much more important than the
functionalization of the cap. Direct covalent sidewall functionalization is
associated with a change of hybridization from sp? to sp* and a
simultaneous loss of conjugation. An alternative approach to covalent
functionalization involves the reaction of defects present (or generated) in
the structure of the nanotube. Defect sites can be the open ends and holes in
the sidewalls, and pentagon and heptagon irregularities in the hexagon
graphene framework (often associated with bends in the tubes). All these
functionalizations are exohedral derivatizations. Taking the hollow structure
of nanotubes into consideration, endohedral functionalization of SWNTs is
possible, i.e., the filling of the tubes with atoms or small molecules. It is
important to note that covalent functionalization methods have one problem
in common: extensive covalent functionalization modifies SWNT
properties by disrupting the continuous m—-system of SWNTSs.

Various applications of nanotubes require different, specific modification to
achieve desirable physical and chemical properties of nanotubes. In this
regard, covalent functionalization provides a higher degree of fine-tuning
the chemistry and physics of SWNTs than non-covalent functionalization.
Until now, a variety of methods have been used to achieve the
functionalization of nanotubes ((link]).

foeee fluorination _____-»_ subsequent reactions
oxidation

azomethine ylides <—__ Li(Na)/Hg

; : alkyl halide
Bingel reaction ; :
carbene and radical reactions

Schematic description of various covalent
functionalization strategies for SWNTs.

Taking chemistry developed for Cgg, SWNTs may be functionalized using
1,3 dipolar addition of azomethine ylides. The functionalized SWNTs are
soluble in most common organic solvents. The azomethine ylide
functionalization method was also used for the purification of SWNTs.
Under electrochemical conditions, aryl diazonium salts react with SWNTs
to achieve functionalized SWNTs, alternatively the diazonium ions may be
generated in-situ from the corresponding aniline, while a solvent free
reaction provides the best chance for large-scale functionalization this way.
In each of these methods it is possible to control the amount of
functionalization on the tube by varying reaction times and the reagents
used; functionalization as high as 1 group per every 10 - 25 carbon atoms is
possible.

Organic functionalization through the use of alkyl halides, a radical
pathway, on tubes treated with lithium in liquid ammonia offers a simple
and flexible route to a range of functional groups. In this reaction,
functionalization occurs on every 17 carbons. Most success has been found
when the tubes are dodecylated. These tubes are soluble in chloroform,
DMF, and THF.

The addition of oxygen moieties to SWNT sidewalls can be achieved by
treatment with acid or wet air oxidation, and ozonolysis. The direct
epoxidation of SWNTs may be accomplished by the direct reaction with a
peroxide reagent, or catalytically. Catalytic de-epoxidation (({link]) allows

for the quantitative analysis of sidewall epoxide and led to the surprising
result that previously assumed “pure” SWNTs actually contain ca. 1 oxygen
per 250 carbon atoms.

ReMeO3 + H909
\ J
ReMeO3 + PPhg

Catalytic oxidation and de-
epoxidation of SWNTs.

One of the easiest functionalization routes, and a useful synthon for
subsequent conversions, is the fluorination of SWNTs, using elemental
fluorine. Importantly, a C:F ratios of up to 2:1 can be achieved without
disruption of the tubular structure. The fluorinated SWNTs (F-SWNTs)
proved to be much more soluble than pristine SWNTs in alcohols (1 mg/mL
in iso-propanol), DMF and other selected organic solvents. Scanning
tunneling microscopy (STM) revealed that the fluorine formed bands of
approximately 20 nm, while calculations using DFT revealed 1,2 addition is
more energetically preferable than 1,4 addition, which has been confirmed
by solid state '3C NMR. F-SWNTs make highly flexible synthons and
subsequent elaboration has been performed with organo lithium, Grignard
reagents, and amines.

Functionalized nanotubes can be characterized by a variety of techniques,
such as atomic force microscopy (AFM), transmission electron microscopy
(TEM), UV-vis spectroscopy, and Raman spectroscopy. Changes in the

Raman spectrum of a nanotube sample can indicate if functionalization has
occurred. Pristine tubes exhibit two distinct bands. They are the radial
breathing mode (230 cm-!) and the tangential mode (1590 cm“!). When
functionalized, a new band, called the disorder band, appears at ca.1350
cm’!. This band is attributed to sp°-hybridized carbons in the tube.
Unfortunately, while the presence of a significant D mode is consistent with
sidewall functionalization and the relative intensity of D (disorder) mode
versus the tangential G mode (1550 — 1600 cm‘) is often used as a measure
of the level of substitution. However, it has been shown that Raman is an
unreliable method for determination of the extent of functionalization since
the relative intensity of the D band is also a function of the substituents
distribution as well as concentration. Recent studies suggest that solid state
13C NMR are possibly the only definitive method of demonstrating covalent
attachment of particular functional groups.

Coating carbon nanotubes: creating inorganic nanostructures

Fullerenes, nanotubes and nanofibers represent suitable substrates for the
seeding other materials such as oxides and other minerals, as well as
semiconductors. In this regard, the carbon nanomaterial acts as a seed point
for the growth as well as a method of defining unusual aspect ratios. For
example, silica fibers can be prepared by a number of methods, but it is
only through coating SWNTs that silica nano-fibers with of micron lengths
with tens of nanometers in diameter may be prepared.

While Cp itself does not readily seed the growth of inorganic materials,
liquid phase deposition of oxides, such as silica, in the presence of
fullerenol, Cgq(OH),, results in the formation of uniform oxide spheres. It
appears the fullerenol acts as both a reagent and a physical point for
subsequent oxide growth, and it is Ceo, or an aggregate of Cgo, that is
present within the spherical particle. The addition of fullerenol alters the
morphology and crystal phase of CaCO3 precipitates from aqueous solution,
resulting in the formation of spherical features, 5-pointed flower shaped
clusters, and triangular crystals as opposed to the usual rhombic crystals. In
addition, the meta-stable vaterite phase is observed with the addition of
Ceo(OH)n-

As noted above individual SWNTs may be obtained in solution when
encased in a cylindrical micelle of a suitable surfactant. These
individualized nanotubes can be coated with a range of inorganic materials.
Liquid phase deposition (LPD) appears to have significant advantages over
other methods such as incorporating surfacted SWNTs into a preceramic
matrix, in situ growth of the SWNT in an oxide matrix, and sol-gel
methods. The primary advantage of LPD growth is that individual SWNTs
may be coated rather than bundles or ropes. For example, SWNTs have
been coated with silica by liquid phase deposition (LPD) using a
Silica/H»SiF, solution and a surfactant-stabilized solution of SWNTs. The
thickness of the coating is dependent on the reaction mixture concentration
and the reaction time. The SWNT core can be removed by thermolysis
under oxidizing conditions to leave a silica nano fiber. It is interesting to
note that the use of a surfactant is counter productive when using MWNTs
and VGFs, in this case surface activation of the nanotube offers the suitable
growth initiation. Pre-oxidation of the MWNT or VGF allows for uniform
coatings to be deposited. The coated SWNTs, MWNTs, and VGFs can be
subsequently reacted with suitable surface reagents to impart miscibility in
aqueous solutions, guar gels, and organic matrixes. In addition to simple
oxides, coated nanotubes have been prepared with minerals such as
carbonates and semiconductors.

Bibliography

e S.M. Bachilo, M. S. Strano, C. Kittrell, R. H. Hauge, R. E. Smalley,
and R. B. Weisman, Science, 2002, 298, 2361.

e D.S. Bethune, C. H. Klang, M. S. deVries, G. Gorman, R. Savoy, J.
Vazquez, and R. Beyers, Nature, 1993, 363, 605.

e J, J. Brege, C. Gallaway, and A. R. Barron, J. Phys. Chem., C, 2007,
111, 17812.

e C. A. Dyke and J. M. Tour, J. Am. Chem. Soc., 2003, 125, 1156.

e Z. Ge, J. C. Duchamp, T. Cai, H. W. Gibson, and H. C. Dorn, J. Am.
Chem. Soc., 2005, 127, 16292.

e L. A. Girifalco, M. Hodak, and R. S. Lee, Phys. Rev. B, 2000, 62,
13104.

e T. Guo, P. Nikolaev, A. G. Rinzler, D. Tomanek, D. T. Colbert, and R.
E. Smalley, J. Phys. Chem., 1995, 99, 10694.

J. H. Hafner, M. J. Bronikowski, B. R. Azamian, P. Nikolaev, A. G.
Rinzler, D. T. Colbert, K. A. Smith, and R. E. Smalley, Chem. Phys.
Lett., 1998, 296, 195.

A. Hirsch, Angew. Chem. Int. Ed., 2002, 40, 4002.

S. Iijima and T. Ichihashi, Nature, 1993, 363, 603.

H. R. Jafry, E. A. Whitsitt, and A. R. Barron, J. Mater. Sci., 2007, 42,
7381.

H. W. Kroto, J. R. Heath, S. C. O’Brien, R. F. Curl, and R. E. Smalley,
Nature, 1985, 318, 162.

F, Liang, A. K. Sadana, A. Peera, J. Chattopadhyay, Z. Gu, R. H.
Hauge, and W. E. Billups, Nano Lett., 2004, 4, 1257.

D. Ogrin and A. R. Barron, J. Mol. Cat. A: Chem., 2006, 244, 267.

D. Ogrin, J. Chattopadhyay, A. K. Sadana, E. Billups, and A. R.
Barron, J. Am. Chem. Soc., 2006, 128, 11322.

R. E. Smalley, Acc. Chem. Res., 1992, 25, 98.

M. M. J. Treacy, T. W. Ebbesen, and J. M. Gibson, Nature, 1996, 381,
678.

E. A. Whitsitt and A. R. Barron, Nano Lett., 2003, 3, 775.

J. Yang and A. R. Barron, Chem. Commun., 2004, 2884.

L. Zeng, L. B. Alemany, C. L. Edwards, and A. R. Barron, Nano Res.,
2008, 1, 72.

Graphene

Introduction

Graphene is a one-atom-thick planar sheet of sp*-bonded carbon atoms that
are densely packed in a honeycomb crystal lattice ({link]). The name comes
from “graphite” and “alkene”; graphite itself consists of many graphene
sheets stacked together.

Idealized structure of a single graphene
sheet.

Single-layer graphene nanosheets were first characterized in 2004, prepared
by mechanical exfoliation (the “scotch-tape” method) of bulk graphite.
Later graphene was produced by epitaxial chemical vapor deposition on
silicon carbide and nickel substrates. Most recently, graphene nanoribbons
(GNRs) have been prepared by the oxidative treatment of carbon nanotubes
and by plasma etching of nanotubes embedded in polymer films.

Physical properties of graphene

Graphene has been reported to have a Young’s modulus of 1 TPa and
intrinsic strength of 130 GP; similar to single walled carbon nanotubes
(SWNTs). The electronic properties of graphene also have some similarity
with carbon nanotubes. Graphene is a zero-bandgap semiconductor.
Electron mobility in graphene is extraordinarily high (15,000 cm?/V.s at
room temperature) and ballistic electron transport is reported to be on
length scales comparable to that of SWNTs. One of the most promising
aspects of graphene involves the use of GNRs. Cutting an individual
graphene layer into a long strip can yield semiconducting materials where
the bandgap is tuned by the width of the ribbon.

While graphene’s novel electronic and physical properties guarantee this
material will be studied for years to come, there are some fundamental
obstacles yet to overcome before graphene based materials can be fully
utilized. The aforementioned methods of graphene preparation are effective;
however, they are impractical for large-scale manufacturing. The most
plentiful and inexpensive source of graphene is bulk graphite. Chemical
methods for exfoliation of graphene from graphite provide the most realistic
and scalable approach to graphene materials.

Graphene layers are held together in graphite by enormous van der Waals
forces. Overcoming these forces is the major obstacle to graphite
exfoliation. To date, chemical efforts at graphite exfoliation have been
focused primarily on intercalation, chemical derivatization, thermal
expansion, oxidation-reduction, the use of surfactants, or some combination
of these.

Graphite oxide

Probably the most common route to graphene involves the production of
graphite oxide (GO) by extremely harsh oxidation chemistry. The methods
of Staudenmeier or Hummers are most commonly used to produce GO, a
highly exfoliated material that is dispersible in water. The structure of GO
has been the subject of numerous studies; it is known to contain epoxide
functional groups along the basal plane of sheets as well as hydroxyl and
carboxyl moieties along the edges ({link]). In contrast to other methods for
the synthesis of GO, the the m-peroxybenzoic acid (m-CPBA) oxidation of

microcrystalline synthetic graphite at room temperature yields graphite
epoxide in high yield, without significant additional defects.

Idealized structure proposed for graphene
oxide (GO). Adapted from C. E. Hamilton,
PhD Thesis, Rice University (2009).

As graphite oxide is electrically insulating, it must be converted by
chemical reduction to restore the electronic properties of graphene.
Chemically converted graphene (CCG) is typically reduced by hydrazine or
borohydride. The properties of CCG can never fully match those of
graphene for two reasons:

1. Oxidation to GO introduces defects.
2. Chemical reduction does not fully restore the graphitic structure.

As would be expected, CCG is prone to aggregation unless stabilized.
Graphene materials produced from pristine graphite avoid harsh oxidation
to GO and subsequent (incomplete) reduction; thus, materials produced are
potentially much better suited to electronics applications.

A catalytic approach to the removal of epoxides from fullerenes and
SWNTs has been applied to graphene epoxide and GO. Treatment of
oxidized graphenes with methyltrioxorhenium (MeReO3, MTO) in the
presence of PPh; results in the oxygen transfer, to form O=PPh3 and allow
for quantification of the C:O ratio.

Homogeneous graphene dispersions

An alternate approach to producing graphene materials involves the use of
pristine graphite as starting material. The fundamental value of such an
approach lies in its avoidance of oxidation to GO and subsequent
(incomplete) reduction, thereby preserving the desirable electronic
properties of graphene. There is precedent for exfoliation of pristine
graphite in neat organic solvents without oxidation or surfactants. It has
been reported that N,N-dimethylformamide (DMF) dispersions of graphene
are possible, but no detailed characterization of the dispersions were
reported. In contrast, Coleman and coworkers reported similar dispersions
using N-methylpyrrolidone (NMP), resulting in individual sheets of
graphene at a concentration of <0.01 mg/mL. NMP and DMF are highly
polar solvents, and not ideal in cases where reaction chemistry requires a
nonpolar medium. Further, they are hygroscopic, making their use
problematic when water must be excluded from reaction mixtures. Finally,
DMF is prone to thermal and chemical decomposition.

Recently, dispersions of graphene has been reported in ortho-
dichlorobenzene (ODCB) using a wide range of graphite sources. The
choice of ODCB for graphite exfoliation was based on several criteria:

1. ODCB is a common reaction solvent for fullerenes and is known to
form stable SWNT dispersions.

2. ODCB is a convenient high-boiling aromatic, and is compatible with a
variety of reaction chemistries.

3. ODCB, being aromatic, is able to interact with graphene via 1-1
stacking.

4. It has been suggested that good solvents for graphite exfoliation should
have surface tension values of 40 — 50 mJ/m?. ODCB has a surface
tension of 36.6 mJ/m?, close to the proposed range.

Graphite is readily exfoliated in ODCB with homogenization and
sonication. Three starting materials were successfully dispersed:
microcrystalline synthetic, thermally expanded, and highly ordered
pyrolytic graphite (HOPG). Dispersions of microcrystalline synthetic
graphite have a concentration of 0.03 mg/mL, determined gravimetrically.
Dispersions from expanded graphite and HOPG are less concentrated (0.02
mg/mL).

High resolution transmission electron microscopy (HRTEM) shows mostly
few-layer graphene (n < 5) with single layers and small flakes stacked on
top ({link]). Large graphitic domains are visible; this is further supported by
selected area electron diffraction (SAED) and fast Fourier transform (FFT)
in selected areas. Atomic force microscope (AFM) images of dispersions
sprayed onto silicon substrates shows extremely thin flakes with nearly all
below 10 nm. Average height is 7 - 10 nm. The thinnest are less than 1 nm,
graphene monolayers. Lateral dimensions of nanosheets range from 100 —
5900 nm.

TEM images of single layer graphene from HOPG
dispersion. (a) monolayer and few layer of graphene
stacked with smaller flakes; (b) selected edge region

from (a), (c) selected area from (b) with FFT inset, (d)
HRTEM of boxed region in (c) showing lattice fringes
with FFT inset. Adapted from C. E. Hamilton, PhD
Thesis, Rice University (2009).

As-deposited films cast from ODCB graphene show poor electrical
conductivity, however, after vacuum annealing at 400 °C for 12 hours the
films improve vastly, having sheet resistances on the order of 60 Q/sq. By

comparison, graphene epitaxially grown on Ni has a reported sheet
resistance of 280 Q2/sq.

Covalent functionalization of graphene and graphite oxide

The covalent functionalization of SWNTs is well established. Some routes
to covalently functionalized SWNTs include esterification/ amidation,
reductive alkylation (Billups reaction), and treatment with azomethine
ylides (Prato reaction), diazonium salts, or nitrenes. Conversely, the
chemical derivatization of graphene and GO is still relatively unexplored.

Some methods previously demonstrated for SWNTs have been adapted to
GO or graphene. GO carboxylic acid groups have been converted into acyl
chlorides followed by amidation with long-chain amines. Additionally, the
coupling of primary amines and amino acids via nucleophilic attack of GO
epoxide groups has been reported. Yet another route coupled isocyanates to
carboxylic acid groups of GO. Functionalization of partially reduced GO by
aryldiazonium salts has also been demonstrated. The Billups reaction has
been performed on the intercalation compound potassium graphite (Cgk),
as well as graphite fluoride, and most recently GO. Graphene alkylation has
been accomplished by treating graphite fluoride with alkyllithium reagents.

ODCB dispersions of graphene may be readily converted to covalently
functionalize graphene. Thermal decomposition of benzoyl peroxide is used
to initiate radical addition of alkyl iodides to graphene in ODCB
dispersions.

Equation:

R-I, benzoyl peroxide R RR R

Additionally, functionalized graphene with nitrenes generated by thermal
decomposition of aryl azides
Equation:

\
ArN3 N

Ar Ar

Bibliography

P. Blake, P. D. Brimicombe, R. R. Nair, T. J. Booth, D. Jiang, F.
Schedin, L. A. Ponomarenko, S. V. Morozov, H. F. Gleeson, E. W.
Hill, A. K. Geim, and K. S. Novoselov, Nano Lett., 2008, 8, 1704.

J. Chattopadhyay, A. Mukherjee, C. E. Hamilton, J.-H. Kang, S.
Chakraborty, W. Guo, K. F. Kelly, A. R. Barron, and W. E. Billups, J.
Am. Chem. Soc., 2008, 130, 5414.

G. Eda, G. Fanchini, and M. Chhowalla, Nat. Nanotechnol., 2008, 3,
270.

M. Y. Han, B. Ozyilmaz, Y. Zhang, and P. Kim, Phys. Rev. Lett., 2008,
98, 206805.

Y. Hernandez, V. Nicolosi, M. Lotya, F. M. Blighe, Z. Sun, S. De, I. T.
McGovern, B. Holland, M. Byme, Y. K. Gun’Ko, J. J. Boland, P.
Niraj, G. Duesberg, S. Krishnamurthy, R. Goodhue, J. Hutchinson, V.
Scardaci, A. C. Ferrari, and J. N. Coleman, Nat. Nanotechnol., 2008, 3,
563.

W. S. Hummers and R. E. Offeman, J. Am. Chem. Soc., 1958, 80,
1339.

L. Jiao, L. Zhang, X. Wang, G. Diankov, and H. Dai, Nature, 2009,
458, 877.

D. V. Kosynkin, A. L. Higginbotham, A. Sinitskii, J. R. Lomeda, A.
Dimiev, B. K. Price, and J. M. Tour, Nature, 2009, 458, 872.

D. Li, M. B. Mueller, S. Gilje, R. B. Kaner, and G. G. Wallace, Nat.
Nanotechnol., 2008, 3, 101.

S. Niyogi, E. Bekyarova, M. E. Itkis, J. L. McWilliams, M. A. Hamon,
and R. C. Haddon, J. Am. Chem. Soc., 2006, 128, 7720.

Y. Si and E. T. Samulski, Nano Lett., 2008, 8, 1679.

L. Staudenmaier, Ber. Dtsch. Chem. Ges., 1898, 31, 1481.

Rolling Molecules on Surfaces Under STM Imaging

Introduction to surface motions at the molecular level

As single molecule imaging methods such as scanning tunneling
microscope (STM), atomic force microscope (AFM), and transmission
electron microscope (TEM) developed in the past decades, scientists have
gained powerful tools to explore molecular structures and behaviors in
previously unknown areas. Among these imaging methods, STM is
probably the most suitable one to observe detail at molecular level. STM
can operate in a wide range of conditions, provides very high resolution,
and able to manipulate molecular motions with the tip. An interesting early
example came from IBM in 1990, in which the STM was used to position
individual atoms for the first time, spelling out "I-B-M" in Xenon atoms.
This work revealed that observation and control of single atoms and
molecular motions on surfaces were possible.

The IBM work, and subsequent experiments, relied on the fact that STM tip
always exerts a finite force toward an adsorbate atom that contains both van
der Waals and electrostatic forces was utilized for manipulation purpose. By
adjusting the position and the voltage of the tip, the interactions between the
tip and the target molecule were changed. Therefore, applying/releasing
force to a single atom and make it move was possible [link].

STM Tip

Metal Substrate

Manipulation of STM tip toward a xenon
atom. a) STM tip move onto a target atom
then change the voltage and current of the
tip to apply a stronger interaction. b) Move

the atom to a desire position. c) After
reaching the desire position, the tip released
by switching back to the scanning voltage
and current.

The actual positioning experiment was carried out in the following process.
The nickel metal substrate was prepared by cycles of argon-ion sputtering,
followed by annealing in a partial pressure of oxygen to remove surface
carbon and other impurities. After the cleaning process, the sample was
cooled to 4 K, and imaged with the STM to ensure the quality of surface.

The nickel sample was then doped with xenon. An image of the doped
sample was taken at constant-current scanning conditions. Each xenon atom
appears as a located randomly 1.6 A high bump on the surface ([link]a).
Under the imaging conditions (tip bias = 0.010 V with tunneling current 10°
9 A) the interaction of the xenon with the tip is too weak to cause the
position of the xenon atom to be perturbed. To move an atom, the STM tip
was placed on top of the atom performing the procedure depicted in [link]
to move it to its target. Repeating this process again and again led the
researcher to build of the structure they desired [link]b and c.

Manipulation of STM tip starting with a) randomly dosed xenon
sample, b) under construction - move xenon atom to desire position,
and c) accomplishment of the manipulation. Adapted from D. M.
Eigler and E. K. Schweizer, Nature, 1990, 344, 524.

All motions on surfaces at the single molecule level can be described as by
the following (or combination of the following) modes:

i. Sliding.
li. Hopping.
iii. Rolling.
iv. Pivoting.

Although the power of STM imaging has been demonstrated, imaging of
molecules themselves is still often a difficult task. The successful imaging
of the IBM work was attributed to selection of a heavy atom. Other
synthetic organic molecules without heavy atoms are much more difficult to
be imaged under STM. Determinations of the mechanism of molecular

motion is another. Besides imaging methods themselves, other auxiliary
methods such as DFT calculations and imaging of properly designed
molecules are required to determine the mechanism by which a particular
molecule moves across a surface.

Herein, we are particularly interested in surface-rolling molecules, i.e.,
those that are designed to roll on a surface. It is straightforward to imagine
that if we want to construct (and image) surface-rolling molecules, we must
think of making highly symmetrical structures. In addition, the magnitudes
of interactions between the molecules and the surfaces have to be adequate;
otherwise the molecules will be more susceptible to slide/hop or stick on
the surfaces, instead of rolling. As a result, only very few molecules are
known can roll and be detected on surfaces.

Surface rolling of molecules under the manipulation of STM
tips

As described above, rolling motions are most likely to be observed on
molecules having high degree of symmetry and suitable interactions
between themselves and the surface. C¢o is not only a highly symmetrical
molecule but also readily imageable under STM due to its size. These
properties together make Cgg and its derivatives highly suitable to study
with regards to surface-rolling motion.

The STM imaging of Cg was first carried out at At King College, London.
Similar to the atom positioning experiment by IBM, STM tip manipulation
was also utilized to achieve Cg, displacement. The tip trajectory suggested
that a rolling motion took into account the displacement on the surface of
Cg. In order to confirm the hypothesis, the researchers also employed ab
initio density function (DFT) calculations with rolling model boundary
condition ({link]). The calculation result has supported their experimental
result.

a) b) Cc)

\
@2@@aAPAea@aAea® @_.@@ ®@ @
GSB@aQ @O2@aAa® @28®

Proposed mechanism of C¢p translation showing the alteration of
Cgo°surface interactions during rolling. a) 2-point interaction.
The left point interaction was dissociated during the interaction.
b) 1-point interaction. Cgg can pivot on surface. c) 2-point
interaction. A new interaction formed to complete part of the
rolling motion. a) - c) The black spot on the Cgg is moved during
the manipulation. The light blue Si balls represent the first layer
of molecules the silicon surface, and the yellow balls are the
second layer.

The results provided insights into the dynamical response of covalently
bound molecules to manipulation. The sequential breaking and reforming of
highly directional covalent bonds resulted in a dynamical molecular
response in which bond breaking, rotation, and translation are intimately
coupled in a rolling motion ({link]), but not performing sliding or hopping
motion.

A triptycene wheeled dimeric molecule [link] was also synthesized for
studying rolling motion under STM. This "tripod-like" triptycene wheel
ulike a ball like Cgg molecule also demonstrated a rolling motion on the
surface. The two triptycene units were connected via a dialkynyl axle, for
both desired molecule orientation sitting on surface and directional
preference of the rolling motion. STM controlling and imaging was
demonstrated, including the mechanism [link].

Scheme of the rolling mechanism (left to right).
Step 1 is the tip approach towards the molecule,
step 2 is a 120 degree rotation of a wheel
around its molecular axle and in step 3 the tip
reaches the other side of the molecule. It shows
that, in principle, only one rotation of a wheel
can be induced (the direction of movement is
marked by arrows).

Single molecule nanocar under STM imaging

Another use of STM imaging at single molecule imaging is the single
molecule nanocar by the Tour group at Rice University. The concept of a
nanocar initially employed the free rotation of a C-C single bond between a
spherical Cgg molecule and an alkyne, [link]. Based on this concept, an
“axle” can be designed into which are mounted Cgp “wheels” connected
with a “chassis” to construct the “nanocar”. Nanocars with this design are

expected to have a directional movement perpendicular to the axle.
Unfortunately, the first generation nanocar (named “nanotruck” [link])
encountered some difficulties in STM imaging due to its chemical
instability and insolubility. Therefore, a new of design of nanocar based on
OPE has been synthesized [link].

Structure of Cg
wheels connecting to
an alkyne. The only
possible rolling
direction is
perpendicular to the
C-C single bond
between Cgg and the
alkyne. The arrow
indicates the
rotational motion of
Ceo.

Rotating axle Planar chassis Spherical wheel

Structure of the nanotruck. No rolling motion was
observed under STM imaging due to its instability,
insolubility and inseparable unreacted Cgy. The double
head arrow indicates the expected direction of nanocar
movement. Y. Shirai, A. J. Osgood, Y. Zhao, Y. Yao, L.
Saudan, H. Yang, Y.-H. Chiu, L. B. Alemany, T. Sasaki,
J.-F. Morin, J. M. Guerrero, K. F. Kelly, and J. M. Tour,
J. Am. Chem. Soc., 2006, 128, 4854. Copyright
American Chemical Society (2006).

Nanocar based on OPE structure. The size of the nanocar is
3.3 nm X 2.1 nm (W x L). Alkoxy chains were attached to
improve solubility and stability. OPE moiety is also separable
from Cgg. The bold double head arrow indicates the expected
direction of nanocar movement. The dimension of nanocar
was 3.3 nm X 2.1 nm which enable direct observation of the
orientation under STM imaging. Y. Shirai, A. J. Osgood, Y.
Zhao, K. F. Kelly, and J. M. Tour, Nano Lett., 2005, 5, 2330.
Copyright American Chemical Society (2005).

The newly designed nanocar was studied with STM. When the nanocar was
heated to ~200 °C, noticeable displacements of the nanocar were observed
under selected images from a 10 min STM experiment [link]. The
phenomenon that the nanocar moved only at high temperature was
attributed their stability to a relatively strong adhesion force between the

fullerene wheels and the underlying gold. The series of images showed both
pivotal and translational motions on the surfaces.

«H>

Translational Motion

Pivotal and translational
movement of OPE based

nanocar. Acquisition time of
one image is approximately 1
min with (a — e) images were
selected from a series spanning
10 min. The configuration of
the nanocar on surface can be
determined by the distances of
four wheels. a) — b) indicated
the nanocar had made a 80°
pivotal motion. b) — e)
indicated translation
interrupted by small-angle
pivot perturbations. Y. Shirai,
A. J. Osgood, Y. Zhao, K. F.
Kelly, and J. M. Tour, Nano
Lett., 2005, 5, 2330. Copyright
American Chemical Society
(2005).

Although literature studies suggested that the Cgg molecule rolls on the
surface, in the nanocar movement studies it is still not possible to
conclusively conclude that the nanocar moves on surface exclusively via a
rolling mechanism. Hopping, sliding and other moving modes could also be
responsible for the movement of the nanocar since the experiment was
carried out at high temperature conditions, making the Cgg molecules more
energetic to overcome interactions between surfaces.

To tackle the question of the mode of translation, a trimeric “nano-tricycle”
has been synthesized. If the movement of fullerene-wheeled nanocar was
based on a hopping or sliding mechanism, the trimer should give observable
translational motions like the four-wheeled nanocar, however, if rolling is
the operable motion then the nano-tricycle should rotate on an axis, but not
translate across the surface. The result of the imaging experiment of the
trimer at ~200 °C ([{link],) yielded very small and insignificant translational
displacements in comparison to 4-wheel nanocar ((link]). The trimeric 3-

wheel nanocar showed some pivoting motions in the images. This motion
type can be attributed to the directional preferences of the wheels mounted
on the trimer causing the car to rotate. All the experimental results
suggested that a Cgp-based nanocar moves via a rolling motion rather than
hopping and sliding. In addition, the fact that the thermally driven nanocar
only moves in high temperature also suggests that four Cgg have very strong
interactions to the surface.

\

Cana

Pivoting Motion

Pivot motion of the
trimer. a) - d) Pivot
motions of circled

trimered were shown in
the series of images. No
significant translation
were observed in
comparison to the
nanocar. Y. Shirai, A. J.
Osgood, Y. Zhao, K. F.
Kelly, and J. M. Tour,
Nano Lett., 2005, 5,
2330. Copyright
American Chemical
Society (2005).

Bibliography

e D. M. Eigler and E. K. Schweizer, Nature, 1990, 344, 524.

e L. Grill, K. -H. Rieder, F. Moresco, G. Rapenne, S. Stojkovic, X.
Bouju, and C. Joachim, Nat. Nanotechnol., 2007, 2, 95.

e Y. Shirai, A. J. Osgood, Y. Zhao, K. F. Kelly, and J. M. Tour, Nano
Lett., 2005, 5, 2330.

e Y. Shirai, A. J. Osgood, Y. Zhao, Y. Yao, L. Saudan, H. Yang, Y.-H.
Chiu, L. B. Alemany, T. Sasaki, J.-F. Morin, J. M. Guerrero, K. F.
Kelly, and J. M. Tour, J. Am. Chem. Soc., 2006, 128, 4854.

The Environmental Impact of the Manufacturing of Seminconductors

This module gives a brief general overview of semi-conductor
manufacturing and some of the components and processes used to produce
them that can potentially cause harm to humans or the environment.

Note:"This module was developed as part of a Rice University Class called
"Nanotechnology: Content and Context" initially funded by the National
Science Foundation under Grant No. EEC-0407237. It was conceived,
researched, written and edited by students in the Fall 2005 version of the
class, and reviewed by participating professors."

What is a semiconductor?

The semiconductor industry is one of the fastest growing manufacturing
sectors in not only the United States but also in the world. According to the
American Electronics Association, the domestic sales of electronic
components have skyrocketed, jumping from $127 billion to $306 billion
over the course of the 1980’s. In the first three quarters of the 2003 fiscal
year alone, the export of technology goods from the United States increased
by $19 billion [1].

The word “semiconductor” technically refers to any member of a class of
solid, crystalline materials that is characterized by an electrical conductivity
better than that of insulators (e.g., plastic) but less than that of good
conductors (e.g., copper) [2]. Semiconductors are particularly useful as a
base material in the manufacturing of computer chips, and the term
semiconductor has actually come to be synonymous with the computer
chips, themselves. However, semiconductors are not only used in
computers. Computers only make up 44% of entire industry consumption
(see [link]). Semiconductors are also used for military, automotive,
industrial, communications, and other consumer purposes.

Military
Automotive 1%
7%

Industnial
8%

Computer
44%
Corsumer
19%

Communications
21%

Relative consumption of semiconductors by
industry [3].

Semiconductors seem to be anywhere and everywhere throughout our
everyday lives, yet it is surprising how little most people know about how
they actually work or about the potentially devastating effects their
manufacturing can have on the environment and human health.

Why is nanotechnology important to the semiconductor
industry?

Much of the study of nanotechnology has been centered on the
manufacturing of semiconductors. Though there are a number of highly
anticipated applications for nanotechnology in other fields, notably in
medicine and in biotechnology, the most tangible results thus far can be
argued to have been achieved in the semiconductor industry.

An example of a semiconductor
(photo from PEAK).

For example, Intel recently unveiled its first products based on a generation
of 90-nanometer process technology, and its researches and engineers have
built and tested prototype transistors all the way down to the 22-nanometer
range. Currently, Intel scientists and engineers are working on identifying
new materials such as carbon nanotubes and nanowires to replace current
transistors, and in particular they hope to develop a “tri-gate” transistor
approach that would enable chip designers to build transistors below the 22-
nanometer range [4].

With the advent of nanotechnology, these transistors are becoming even
faster and more powerful, and in accordance with the law of accelerating
returns, the industry has been producing smaller transistors at lower costs
with each and every passing year. As these semiconductors become smaller
and smaller, they are quickly and surely pushing towards the limits of the
nano-realm.

These innovations, however, do not come without their fair share of
challenges. Physical issues that are not problematic at the micron scale arise
at the nano-scale due to the emergence of quantum effects, and in much the

same way that optical microscopy cannot be utilized at the nano-scale, the
semiconductor industry is fast approaching a similar diffraction limit.
Optical lithography, for instance, a process that uses the properties of light
to etch transistors onto wafers of silicon, will soon reach its limit.

At its most basic level, nanotechnology involves pushing individual atoms
together one by one. Since approximately 1.7 billion transistors are required
for a single chip, this is obviously not a realistic method for mass
production. Unless an alternative method for production or a solution to this
problem is found, the development and manufacturing of transistors are
expected to hit a proverbial brick wall by the year 2015. This is the reason
that research in nanotechnology is so important for the world and future of
semiconductors.

How are semiconductors manufactured?

Today’s semiconductors are usually composed of silicon, and they are
manufactured in a procedure that combines the familiar with the bizarre;
some steps that are involved in the process are as everyday as developing a
roll of photographic film while others seem as if they would be better suited
to take place on a spaceship.

These semiconductors appear to the naked eye as being small and flat, but
they are actually three-dimensional “sandwiches” that are ten to twenty
layers thick. It can take more than two dozen steps and up to two full
months to produce a single one of these silicon sandwiches. Some of the
basic and more essential steps involved in the manufacturing process of
silicon chips are briefly detailed below.

First, silicon crystals are melted in a vat and purified to 99.9999% purity.
The molten silicon is drawn into long, heavy, cylindrical ingots, which are
then cut into thin slices called wafersabout the thickness of a business card.

One side of each wafer must be polished absolutely smooth. This process is
called chemical-mechanical polishing, and it involves bathing the wafers in
special abrasive chemicals. After chemical-mechanical polishing,

imperfections cannot be detected on the wafers even with the aid of a
laboratory microscope.

After a wafer is polished, layers of material must be stacked on top of the
silicon wafer base. Insulating layers are laid down in alternation with
conducting layers in a process called deposition. This is often achieved by
spraying the chemicals directly onto the surface of the wafer through
chemical vapor deposition. Following deposition, the wafer is coated with
another layer of chemicals called a photoresist that is sensitive to light.

Next, a machine called a stepper ({link]) is calibrated to project an
extremely fine and focused image through a special type of reticle film in a
manner similar to that of a simple slide projector. The light that is
transmitted through the reticle is projected onto the photoresist layer, which
reacts to the light and begins to harden. All of the parts of the wafer
exposed to this light harden into a tough crust while the parts in shadow
remain soft. This particular step is known by the name of
photoelectrochemical etching because it achieves an etching effect,
resulting in a chip.

(ASM Lithography)

An artist’s illustration of a stepper (image

from Solid State Electronics).

Hundreds of copies of the chip are etched onto the wafer until the entire
surface has been exposed. Once this process is complete, the entire wafer is
submerged into an etching bath, which washes away any parts of the
photoresist that remain unexposed along with the insulating chemicals
underneath. The hardened areas of the photoresist, however, remain and
protect the layers of material underneath them. This process of depositing
chemicals, coating with a photoresist, exposure to light over a film mask,
and etching and washing away is repeated more than a dozen times. The
result is an elaborate, three-dimensional construction of interlocking silicon
wires.

This product is then coated with another insulating layer and is plated with
a thin layer of metal, usually either aluminum or copper. Yet another
photoresist is laid down on top of this metal plating, and after the wafer is
exposed in a stepper, the process repeats with another layer of metal. After
this step has been repeated several more times, a final wash step is
performed, and a finished semiconductor product rolls off the assembly
line, at last.

What is a clean room?

A typical semiconductor fabrication facility, or “fab” in industry jargon,
looks like a normal two- or three-story office building from the outside, and
most of the interior space is devoted to one or more “clean rooms,” in
which the semiconductors are actually made. A clean room is designed with
a fanatical attention to detail aimed towards keeping the room immaculate
and dust-free ({link]).

An industry clean room at AP Tech (photo
from Napa Gateway).

Most if not all surfaces inside these clean rooms are composed of stainless
steel, and these surfaces are sloped whenever possible or perforated by
grating to avoid giving dust a place to settle. The air is filtered through both
the ceiling and the floor to remove particles that are down to 1/100 the
width of a human hair. Lighting is characteristically bright and slightly
yellowish to prevent mildew from forming behind equipment or in recessed
comers, and even the workers in a clean room must be absolutely spotless.

Workers in these rooms must be covered from head to toe in “bunny suits”
that completely seal the body in a bulky suit, helmet, battery pack, gloves,
and boots. Once sealed in these suits, the workers often look more like
space explorers in a science fiction movie than computer chip employees,
but in order to even enter the stainless steel locker room to suit up to begin
with, they must first pass through a series of air lock doors, stand under a
number of “air showers” that actually blow dust off of clothing, and walk
across a sticky floor matting that removes grime from the bottom of shoes.

Semiconductor-manufacturing companies often portray their fabrication
facilities as being clean, environmentally friendly, and conspicuously free
of the black, billowing smokestacks that have come to be associated with

the plants and factories of other major industries. These facilities produce
no visible pollution and certainly do not appear to pose any health or
environmental risks.

In truth, the term “clean room,” itself is more than just a bit of an
understatement. Industry executives often boast that their clean rooms are
from 1,000 times to 10,000 times cleaner and more sanitary than any
hospital operating room.

What are the health risks involved in the semiconductor
industry?

The use of sterile techniques and the fastidious attention devoted to
cleanliness in the semiconductor industry may perpetuate the illusion that
the manufacturing of semiconductors is a safe and sterile process. However,
as arapidly growing body of evidence continues to suggest, hardly anything
could be further from the truth ({link]). The question of worker safety and
chemical contamination at chip-making plants has received an increasing
amount of attention over the course of the past decade.

4. DANGER

TOXIC CHEMICAL HAZARD

WEAR RUBBER DONT BREATHE DON'T INGEST
GLOVES VAPOR CHEMICAL

EXPOSURE TO CHEMICAL CAN RESULT IN SERIOUS
INJURY OR DEATH

FARSHA

Chemicals used in the
manufacturing of semiconductors
are known to have toxic effects
(image from FARSHA).

The devices being built at semiconductor fabrication facilities are super-
sensitive to environmental contaminants. Because each chip takes dozens of
trained personnel several weeks to complete, an enormous amount of time
and effort is expended to produce a single wafer. The industry may pride
itself on its perfectly immaculate laboratories and its bunny-suited workers,
but it should be noted that the bunny suits are not designed to protect their
wearers from hazardous materials but rather to protect the actual
semiconductor products from coming into contact with dirt, hair, flakes of
skin, and other contaminants that can be shed from human bodies. They
protect the silicon wafers from the people, not the people from the
chemicals.

Lee Neal, the head of safety, health, and environmental affairs for the
Semiconductor Industry Association, has been quoted as saying, “This is an
environment that is cleaner than an operating room at a hospital.” However,
this boast is currently being challenged by industry workers, government
scientists, and occupational-health experts across the country and
worldwide.

Industrial hygiene has always been an issue in the semiconductor industry.
Many of the chemicals involved in the manufacturing process of
semiconductors are known human carcinogens or pose some other serious
health risk if not contained properly. [link] lists ten of the hazardous
chemicals most commonly used in manufacturing semiconductors along
with their known effects on human health.

Chemical name

Acetone

Arsenic

Arsine

Benzene

Role in
manufacturing
process

Chemical-
mechanical polishing
of silicon wafers

Increases
conductivity of
semiconductor
material

Chemical vapor
deposition

Photoelectrochemical

Health problems
linked to exposure

Nose, throat, lung,
and eye irritation,
damage to the skin,
confusion,
unconsciousness,
possible coma

Nausea, delirium,
vomiting,
dyspepsia, diarrhea,
decrease in
erythrocyte and
leukocyte
production,
abnormal heart
rhythm, blood
vessel damage,
extensive tissue
damage to nerves,
stomach, intestine,
and skin, known
human carcinogen
for lung cancer

Headache, malaise,
weakness, vertigo,
dyspnea, nausea,
abdominal and
back pain, jaundice,
peripheral
neuropathy, anemia

Damage to bone

etching

Creates “holes” in
silicon lattice to

Cadmium
create effect of
positive charge
Hydrochloric Photoelectrochemical
acid etching
Lead Electroplated

soldering

Marrow, anemia,
excessive bleeding,
immune system
effects, increased
chance of infection,
reproductive
effects, known
human carcinogen
for leukemia

Damage to lungs,
renal dysfunction,
immediate hepatic
injury, bone
defects,
hypertension,
reproductive
toxicity,
teratogenicity,
known human
carcinogen for lung
and prostate cancer

Highly corrosive,
severe eye and skin
burns,
conjunctivitis,
dermatitis,
respiratory
irritation

Damage to renal,
reproductive, and
immune systems,
spontaneous
abortion, premature
birth, low birth

Methyl
chloroform Wasting
Toluene Chemical vapor

deposition

Trichloroethylene Washing

weight, learning
deficits in children,
anemia, memory
effects, dementia,
decreased reaction
time, decreased
mental ability

Headache, central
nervous system
depression, poor
equilibrium, eye,
nose, throat, and
skin irritation,
cardiac arrhythmia

Weakness,
confusion, memory
loss, nausea,
permanent damage
to brain, speech,
vision, and hearing
problems, loss of
muscle control,
poor balance,
neurological
problems and
retardation of
growth in children,
suspected human
carcinogen for lung
and liver cancer

Irritation of skin,
eyes, and
respiratory tract,
dizziness,

drowsiness, speech
and hearing
impairment, kidney
disease, blood
disorders, stroke,
diabetes, suspected
human carcinogen
for renal cancer

Chemicals of concern in the semiconductor industry [5].

Several semiconductor manufacturers including National Semiconductor
and IBM have been cited in the past for holes in their safety procedures and
have been ordered to tighten their handling of carcinogenic and toxic
materials.

In 1996, 117 former employees of IBM and the families of 11 workers who
had died of cancer filed suit against the chemical manufacturers Eastman
Kodak Company, Union Carbide Corporation, J. T. Baker, and KTI
Chemicals, claiming that they had suffered adverse health effects as a result
of exposure to hazardous chemicals on the job in the semiconductor
industry [5]. The lawsuit was filed in New York, which prevented the
employees from suing IBM directly. A separate group of former IBM
workers who had developed cancer filed suit against the company in
California, alleging that they had been exposed to unhealthy doses of
carcinogenic chemicals over the past three decades. Witnesses who testified
in depositions in the New York state court in Westchester County described
how monitors that were supposed to warn workers of toxic leaks often did
not function because of corrosion from acids and water. They also alleged
that supervisors sometimes shut down monitors to maintain production
rates. When they lodged complaints with senior officials in the company,
they claim to have been told not to “make waves” [6]. Meanwhile, 70
female workers in Scotland sued National Semiconductor Corporation,
another U.S.-based company, claiming that they, too, were exposed to
carcinogens on the job.

These lawsuits and the resulting publicity prompted a groundbreaking study
by the Health and Safety Executive, which commissioned a committee to
investigate these allegations [7]. The committee found that there were
indeed unusually high levels of breast and other kinds of cancer among
workers at National Semiconductor’s fabrication facility in Greenock,
Scotland. The committee concluded that the company had failed to ensure
that the local exhaust ventilation systems adequately controlled the potential
exposure of employees to hydrofluoric acid and sulphuric acid fumes and to
arsenic dust. These findings proved to be extremely embarrassing for the
company and for the industry. According to an official statement released
by Ira Leighton, acting regional administrator of the New England branch
of the U.S. Environmental Protection Agency, "National Semiconductor is a
big business that uses a large amount of harmful chemicals and other
materials. Our hazardous waste regulations were created to properly
monitor dangerous chemicals and prevent spills. In order for it to work, it is
important businesses to comply with all of the regulations. When
companies fail to do this they are potentially putting people — their
employees and neighbors — at risk [8]. "

Moreover, a study of fifteen semiconductor manufacturers published in the
December 1995 issue of the American Journal of Independent Medicine
showed that women working in the so-called clean rooms of the
semiconductor fabs suffered from a 14% miscarriage rate.

Protesters at a rally staged against
IBM (photo from San Francisco
Independent Media Center).

The main problem in prosecution is that the industry does not have a single
overarching and definitive process for manufacturing, and it is difficult to
pinpoint one particular compound as causing a certain health problem
because some plants use as many as 300 chemicals. Also, many of the
manufacturing processes take place in closed systems, so exposure to
harmful substances is often difficult to detect unless monitored on a daily
basis.

Executives and spokespeople for the semiconductor industry maintain that
any chip workers’ cancers and other medical problems are more likely due
to factors unrelated to the job, such as family history, drinking, smoking, or
eating habits. They also say that over the years, as awareness of chemical
hazards has grown, they have made efforts to phase out toxic chemicals and
to lower exposure to others. They insist that they use state-of-the-art
process equipment and chemical transfer systems that limit or prevent
physical exposure to chemicals and point out that the substances used in the
semiconductor industry are used in other industries without a major health
or safety problem.

What environmental risks are involved?

In theory, attention to cleanliness is in the manufacturer’s best interest not
only from a health perspective but also from an economic. Many chemicals
used in the production process are not expensive in and of themselves;
however, the cost of maintaining these materials in an ultra-clean state can
be quite high. This encourages the close monitoring of usage, the
minimization of consumption, and the development of recycling and
reprocessing techniques. Also, the rising costs of chemical disposal are
prompting companies to conduct research into alternatives that use more
environmentally friendly methods and materials. Individual companies and

worldwide trade associations were active in reducing the use and emission
of greenhouse gases during the 1990’s, and the industry as a whole has
substantially reduced emissions over the last twenty years.

Nonetheless, there has been a history of environmental problems linked to
the industry in Silicon Valley and other technology centers. To begin with, a
tremendous amount of raw materials is invested in the manufacturing of
semiconductors every year.

Moreover, a typical facility producing semiconductors on six-inch wafers
reportedly uses not only 240,000 kilowatt hours of electricity but also over
2 million gallons of water every day [9]. Newer facilities that produce
eight-inch and twelve-inch wafers consume even more, with some estimates
going as high as five million gallons of water daily. While recycling and
reusing of water does occur, extensive chemical treatment is required for
remediation, and in dry or desert areas such as Albuquerque, New Mexico,
home to plants for Motorola, Philips Semiconductor, Allied Signal and
Signetics, Intel, and other high-tech firms, the high consumption of water
necessary for the manufacturing of semiconductors can pose an especially
significant drain on an already scarce natural resource [10]. The existence
of economic mainstays including the mining industry and the established
presences of Sandia National Laboratories and the Los Alamos National
Laboratory make New Mexico an attractive location for high-tech tenants.
However, the opening of fabrication facilities in the state leaves its farmers
and ranchers in constant competition with the corporations for rights to
water consumption. On average, the manufacturing of just 1/8-inch of a
silicon wafer requires about 3,787 gallons of wastewater, not to mention 27
pounds of chemicals and 29 cubic feet of hazardous gases [11].

A community near Sutter Creek,
California that has been designated as
an EPA Superfund site as a result of
arsenic contamination (photo from
Alexander, Hawes, & Audet).

Contamination has also been an issue in areas surrounding fabrication
plants. Drinking water was found to be contaminated with trichloroethane
and Freon, toxins commonly used in the semiconductor industry, in San
Jose, California in 1981 [12]. These toxins were later suspected to be the
cause of birth defects of many children in the area. The culprits were
Fairchild Semiconductor and IBM. The companies’ underground storage
tanks were found to have leaked tens of thousands of gallons of the toxic
solvents into the ground. There are a number of semiconductor-related EPA
cleanup sites in Silicon Valley, and there have been concerns raised about
the cumulative air and groundwater pollution in Silicon Valley, as well.

Another area of concern is the eventual fate of discarded electronic systems
such as computers, pagers, mobile phones, and televisions that contain
semiconductor devices. Personal computers in particular are especially
problematic because they become obsolete fairly rapidly and lose almost all
of their market value within five or ten years after their date of
manufacture. Tens of millions of PC’s are sold in the United States each
year, and they pose an environmental risk not only through their sheer bulk

in city dumps and landfills but also because their semiconducting devices
often contain significant amounts of heavy metals, including lead and other
potentially hazardous substances.

Why don’t we hear more about this on the news?

Across the United States, approximately 60% of the manufacturing
facilities for semiconductor devices are located in six states. These states
listed in descending order are California, Texas, Massachusetts, New York,
Illinois, and Pennsylvania. The industry appears to be concentrated in these
particular locations in part because they are near the primary users,
transportation routes, and experts in the field, but people of all ages in all
fifty states are impacted by semiconductor technology. Consumerism of
semiconductor products is only expected to increase in coming years.
Apple, for instance, expects to have sold 23.6 million iPods, devices that
rely on semiconductor technology, by the year 2006.

If semiconductors are so ubiquitous in our day-to-day lives, why is there so
little awareness about the serious environmental and health risks that are
involved in their manufacturing process? Part of the problem is that little is
known about the long-term health or environmental consequences of
exposure to the chemicals that are used in the process. Because the
semiconductor industry is still relatively new, not many studies have been
conducted on this topic, and existing data is often inconclusive. This being
said, some scientists predict that the cancer rate in the silicon chip industry
will rise significantly in the future because cancer can take as long as 20-25
years to manifest itself in populations of exposed workers.

The EPA does have regulations in effect that are aimed toward the purpose
of controlling the levels of contaminants released and minimizing human
and environmental exposure to them. However, current regulations do not
mandate that American companies report on offshore manufacturing.
Therefore, even as media coverage and general awareness increase,
companies can simply outsource more and more of their fabrication
facilities to, for example, Southeast Asia. Some companies, in fact, have
begun to do so, and there have even already been studies conducted on the

health issues of workers in the electronics and semiconductor industries of
Singapore and Malaysia [13].

Thus, changes in how and where semiconductor firms manufacture chips
currently outstrip the present ability of the United States government and
media institutions to track and monitor their potential threats to humans and
the environment. If this situation is to change for the better in the near
future, it is clear that radical reforms will need to take place on a number of
different levels. However, the who, what, when, where, and why, so to
speak, of that reform remains to be addressed.

Discussion questions

How many electronics products do you use on a day-to-day basis?
How many of these products contain semiconductors?

Who do you think is ultimately responsible for initiating reform? The
government? The corporation? The consumer?

Do you think that the health and environmental incidents related to
semiconductor manufacturing will remain isolated incidents? Or do
you think that these incidents will become epidemic in the future?
Do you think that nanotechnology will help the problem or make the
problem worse?

Endnotes

1. M. Kazmierczak and J. James. Industry Data & Publications: U.S.

High-Tech Exports, 2000-2004. 16 Nov. 2004. American Electronics
Association. 17 Oct.
2005<http://www.aeanet.org/Publications/idjl_ushightechexports1204.
asp>.

. J. Turley, The Essential Guide to Semiconductors. Upper Saddle River,

New Jersey: Prentice Hall Professional Technical Reference, 2003.

. J. Turley, The Essential Guide to Semiconductors. Upper Saddle River,

New Jersey: Prentice Hall Professional Technical Reference, 2003.
From Prentice Hall

. IBM Research Nanotechnology Homepage. IBM. 16 Oct. 2005

<http://domino.research.ibm.com/comm/research.nsf/pages/r.nanotech.

10.

td.

2.

13:

html>.

. R. Chepesiuk, “Where the Chips Fall: Environmental Health in the

Semiconductor Industry.” Environmental Health Perspectives 107
(1999): 452-457.

. Richards, “Industry Challenge: Computer-Chip Plants Aren’t as Safe

And Clean As Billed, Some Say — Women at Scottish Factory Tell of
Spills and Fumes, Face Host of Medical Ills — Firms Won’t Help Do a
Study.” Wall Street Journal 5 Oct. 1998, eastern ed.: A1.

. A. Heavens, Chip Plants Take Heat For Toxics. 14 Jan. 2003. Wired

News. 13 Oct. 2005,
<http://www. wired.com/news/technology/0,1282,57191,00.html>

. M. Merchant, Maine Semiconductor Plant Fined For Hazardous Waste

Violations. Boston: U.S. Environmental Protection Agency, Press
Office, 2001.

. P. Dunn, Cleanliness Outside, Some Issues Outside. 2 Oct. 2000. The

Foundation for American Communications. 13 Oct. 2005
<http://www.facsnet.org/tools/sci_tech/tech/community/environ2.php3
>

J. Mazurek, Making Microchips: Policy, Globalization, and Economic
Restructuring in the Semiconductor Industry. Cambridge,
Massachusetts: MIT Press, 1999.

C. Hayhurst, “Toxic Technology: Electronics and the Silicon Valley.”
E: the Environmental Magazine May-Jun. 1997: 4.

B. Pimentel, “The Valley’s Toxic History.” San Francisco Chronicle 30
Jan. 2004, final ed.: B1.

V. Lin, Health, Women’s Work, and Industrialization: Semiconductor
Workers in Singapore and Malaysia. New York: Garland Publishing,
Ine:,, 1991.